© 2007 jose alejandro aparicio - university of...
TRANSCRIPT
1
VISUAL QUANTIFICATION OF NON-HOMOGENEOUS COLORS IN FOODS
By
JOSE ALEJANDRO APARICIO
A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE
UNIVERSITY OF FLORIDA
2007
2
© 2007 Jose Alejandro Aparicio
3
To Caroline Elizabeth Fisher for your never ending support and encouragement throughout this journey, and who made this milestone possible
4
ACKNOWLEDGMENTS
I am very grateful to my major advisor, Dr. Murat O. Balaban, for his guidance and
support. My appreciation also to the members of my supervisory committee, Dr. Charles Sims
and Dr. Allen Wysocki, for their mentoring, all participants in my surveys for their input and
open participation, and my lab mates for their support. I thank my family for their loyal
encouragement, which always gave me strength to complete my study.
5
TABLE OF CONTENTS page
ACKNOWLEDGMENTS ...............................................................................................................4
LIST OF TABLES...........................................................................................................................7
LIST OF FIGURES .........................................................................................................................9
ABSTRACT...................................................................................................................................13
CHAPTER
1 INTRODUCTION ..................................................................................................................15
2 LITERATURE REVIEW .......................................................................................................17
Color of Foods and Agricultural Materials.............................................................................17 Instrumental Color Measurement in Agricultural Food Products ..........................................17 Computer Vision or Machine Vision System.........................................................................18 Bakery Products......................................................................................................................20 Red Meat and Seafood............................................................................................................20 Vegetables...............................................................................................................................23 Fruits .......................................................................................................................................23 Prepared Consumer Foods......................................................................................................24 Food Container Inspection......................................................................................................24 Grains......................................................................................................................................25 Other Applications..................................................................................................................25 Visual Texture Analysis .........................................................................................................26 Visual Texture Applications in Agriculture ...........................................................................27 Correlation between Image and Visual Color Analysis .........................................................27 Preliminary Experiments ........................................................................................................28 Objectives of the Study...........................................................................................................30
3 MATERIALS AND METHODS ...........................................................................................31
Mangos and Nectarines...........................................................................................................31 Image Acquisition...................................................................................................................32 Image Analysis .......................................................................................................................32 Experimental Design ..............................................................................................................33 Method of Selection of the Reference Color Bars..................................................................34 Sensory Evaluations................................................................................................................35 Determination of Color Uniformity of Fruit...........................................................................36
Average Color: ................................................................................................................36 Color Blocks....................................................................................................................37 Color Primitives...............................................................................................................37
Calculation of Best Possible ΔE .............................................................................................39
6
Statistical Analysis..................................................................................................................39
4 RESULTS & DISCUSSION ..................................................................................................41
MV Color Results of Fruits ....................................................................................................41 Non-Uniformity Analysis of Fruits ........................................................................................42 Best Possible ΔE.....................................................................................................................44 Sensory Panel Results.............................................................................................................48 Statistical Analysis..................................................................................................................50
Mangos ............................................................................................................................51 Nectarines ........................................................................................................................56
ΔE vs. CCI ..............................................................................................................................61
5 CONCLUSIONS ....................................................................................................................62
APPENDIX
A COLOR ANALYSIS FOR ALL TRAYS .................................................................................64
B PANELISTS PERFORMANCE FOR MANGOS AND NECTARINES .................................75
C ΔE VS CCI FOR ALL COMBINATIONS................................................................................79
D DELTA E VALUES FOR DIFFERENT CASES ....................................................................85
E SOURCE CODES FOR SAS PROGRAMS..............................................................................91
LIST OF REFERENCES...............................................................................................................98
BIOGRAPHICAL SKETCH ....................................................................................................105
7
LIST OF TABLES
Table page 3-1 Nikon D200 settings ..........................................................................................................32
3-2 Factorial-Level combinations ............................................................................................34
4-1 MV color analysis for mangos...........................................................................................41
4-2 MV color analysis for nectarines .......................................................................................41
4-3 Best possible selections and minimum ΔE value possible for 8 references and 2 selections for mangos.........................................................................................................45
4-4 Best possible selections and minimum ΔE value possible for 8 references and 2 selections for nectarines.....................................................................................................45
4-5 Best possible selections and minimum ΔE value possible for 12 references and 2 selections for mangos.........................................................................................................45
4-6 Best possible selections and minimum ΔE value possible for 12 references and 2 selections for nectarines.....................................................................................................46
4-7 Best possible selections and minimum ΔE value possible for 16 references and 2 selections for mangos.........................................................................................................46
4-8 Best possible selections and minimum ΔE value possible for 16 references and 2 selections for nectarines.....................................................................................................46
4-9 Best possible selections and minimum ΔE value possible for 8 references and 4 selections for mangos.........................................................................................................47
4-10 Best possible selections and minimum ΔE value possible for 8 references and 4 selections for nectarines.....................................................................................................47
4-11 Best possible selections and minimum ΔE value possible for 12 references and 4 selections for mangos.........................................................................................................47
4-12 Best possible selections and minimum ΔE value possible for 12 references and 4 selections for nectarines.....................................................................................................48
4-13 Summary performance for panelists evaluating mangos for booth 1 ................................49
4-14 Summary performance for panelists evaluating nectarines for booth 1 ............................50
4-15 ANOVA summary absolute ΔE for mangos......................................................................51
8
4-16 ANOVA summary difference ΔE for mangos ...................................................................51
4-17 ANOVA summary absolute ΔE for nectarines ..................................................................56
4-18 ANOVA summary difference in ΔE for nectarines ...........................................................57
A-1 L*a*b values for reference color bar with 8 color.............................................................70
A-2 L*a*b values for reference color bar with 12 colors .........................................................71
A-3 L*a*b values for reference color bar with 16 colors .........................................................71
A-4 Mango color primitives......................................................................................................73
A-5 Nectarine color primitives..................................................................................................74
B-1 Summary performance for panelists evaluating both fruits for booth 1 ............................75
B-2 Summary performance for panelists evaluating both fruits for booth 2 ............................76
B-3 Summary performance for panelists evaluating both fruits for booth 3 ............................76
B-4 Summary performance for panelists evaluating both fruits for booth 4 ............................76
B-5 Summary performance for panelists evaluating both fruits for booth 5 ............................77
B-6 Summary performance for panelists evaluating both fruits for booth 6 ............................77
B-7 Summary performance for panelists evaluating both fruits for booth 7 ............................77
B-8 Summary performance for panelists evaluating both fruits for booth 8 ............................78
B-9 Summary performance for panelists evaluating both fruits for booth 9 ............................78
B-10 Summary performance for panelists evaluating both fruits for booth 10 ..........................78
E-1 Mixed mode summary absolute ΔE for mangos ................................................................91
E-2 Mixed mode summary difference ΔE for mangos .............................................................92
E-3 Mixed Mode summary absolute ΔE for nectarines............................................................92
E-4 Mixed Mode summary difference in ΔE for nectarines.....................................................92
9
LIST OF FIGURES
Figure page 3-1 Example of mango and nectarine on aluminum tray .........................................................31
3-2 Example of reference color bar with 8 colors added to fruit images presented to the panelists..............................................................................................................................35
4-1 Correlation between number of primitives and color change index (CCI)........................43
4-2 Correlation between number of neighbors and color change index (CCI) ........................43
4-3 Correlation between number of neighbors and number of primitives ...............................44
4-4 Comparison of ΔE values for 8, 12, and 16 reference colors, 2 selections........................49
4-4 Absolute ΔE means difference of selections of colors using mangos ...............................52
4-5 Difference in ΔE means difference of selections of colors using mangos.........................52
4-6 Absolute ΔE means for reference colors for mangos.........................................................53
4-7 Difference in ΔE means for reference colors for mangos..................................................53
4-8 Absolute ΔE means for interaction between the number of reference colors and the number of selections ..........................................................................................................54
4-9 Difference in ΔE means for interaction between the number of reference colors and the number of selections ....................................................................................................55
4-10 Absolute ΔE means for presentation for mangos...............................................................55
4-11 Difference ΔE means for presentation for mangos ............................................................56
4-12 Absolute ΔE means for reference colors for nectarines.....................................................57
4-13 Difference ΔE means for reference colors for nectarines ..................................................58
4-14 Absolute ΔE means for selection of colors for nectarine...................................................58
4-15 Difference ΔE means for selections of colors for nectarines.............................................59
4-16 Difference ΔE means for selections of colors for nectarines.............................................59
4-17 Difference ΔE means for selections of colors for nectarines.............................................60
4-18 Absolute ΔE means for presentation for nectarines...........................................................60
10
4-19 Difference ΔE means for presentation for nectarines ........................................................61
A-1 Fruit tray booth 1 for image acquisition and sensory panel...............................................64
A-2 Fruit tray booth 2 for image acquisition and sensory panel...............................................64
A-3 Fruit tray booth 3 for image acquisition and sensory panel...............................................65
A-4 Fruit tray booth 4 for image acquisition and sensory panel...............................................65
A-5 Fruit tray booth for image acquisition and sensory panel.................................................66
A-6 Fruit tray booth 6 for image acquisition and sensory panel...............................................66
A-7 Fruit tray booth 7 for image acquisition and sensory panel...............................................67
A-8 Fruit tray booth 8 for image acquisition and sensory panel...............................................67
A-9 Fruit tray booth 9 for image acquisition and sensory panel...............................................68
A-10 Fruit tray booth 10 for image acquisition and sensory panel.............................................68
A-11 Machine Vision set-up .......................................................................................................69
A-12 Light box specifications.....................................................................................................69
A-13 Reference scales presented to panelists. ............................................................................70
A-14 Example ballot for screen image evaluation......................................................................72
A-15 Example ballot for fruit evaluation. ...................................................................................73
A-16 Representation of color primitives and equivalent circles for mangos (left) and nectarines (right) with a MV system..................................................................................74
C-1 Absolute Δ E for nectarine for screen image and 8 references ..........................................79
C-2 Absolute Δ E for nectarine for screen image and 12 references ........................................79
C-3 Absolute Δ E for nectarine for screen image and 16 references ........................................80
C-4 Absolute Δ E for nectarine for tray and 8 references.........................................................80
C-5 Absolute Δ E for nectarine for tray and 12 references.......................................................81
C-6 Absolute Δ E for nectarine for tray and 16 references.......................................................81
C-7 Absolute Δ E for mango for screen image and 8 references..............................................82
11
C-8 Absolute Δ E for mango for screen image and 12 references............................................82
C-9 Absolute Δ E for mango for screen image and 16 references............................................83
C-10 Absolute Δ E for mango for tray 8 references ...................................................................83
C-11 Absolute Δ E for mango for tray and 12 references...........................................................84
C-12 Absolute Δ E for mango for tray and 16 references...........................................................84
D-1 Absolute Δ E for nectarine for screen image and 8 references ..........................................85
D-2 Absolute Δ E for nectarine for screen image and 12 references ........................................85
D-3 Absolute Δ E for nectarine for screen image and 16 references ........................................86
D-4 Absolute Δ E for nectarine for tray and 8 references.........................................................86
D-5 Absolute Δ E for nectarine for tray and 12 references.......................................................87
D-6 Absolute Δ E for nectarine for tray and 16 references.......................................................87
D-7 Absolute Δ E for mango for screen image and 8 references..............................................88
D-8 Absolute Δ E for mango for screen image and 12 references............................................88
D-9 Absolute Δ E for mango for screen image and 16 references............................................89
D-10 Absolute Δ E for mango for tray 8 references ...................................................................89
D-11 Absolute Δ E for mango for tray and 12 references...........................................................90
D-12 Absolute Δ E for mango for tray and 16 references...........................................................90
E-1 Absolute ΔE means for selection of color for mangos ......................................................93
E-2 Difference ΔE Means for selection of color for mangos ...................................................93
E-3 Absolute ΔE means for reference colors for mangos.........................................................94
E-4 Difference ΔE means for reference colors for mangos......................................................94
E-5 Absolute ΔE means for presentation for mangos...............................................................94
E-6 Difference ΔE means for presentation for mangos ............................................................95
E-7 Absolute ΔE means for selections of colors for nectarines................................................95
12
E-8 Difference ΔE means for selections of colors for nectarines.............................................96
E-9 Absolute ΔE means for reference colors for nectarines.....................................................96
E-11 Absolute ΔE means for presentation for nectarines...........................................................97
E-12 Difference ΔE means for presentation for nectarines ........................................................97
13
Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Science
VISUAL QUANTIFICATION OF NON-HOMOGENEOUS COLORS IN FOODS
By
Jose Alejandro Aparicio
December 2007
Chair: Murat Balaban Major: Food Science and Human Nutrition
Color is an important quality attribute for nearly every agricultural product. Consumers
may perceive color as an indicator of freshness and wholesomeness, and color may affect their
final decision to accept/reject food. A better understanding of human perception of colors in
food would be beneficial to increase the consistency and quality of food products. The
quantification of color is becoming more important due to an emphasis on international trade and
implementation of Hazard Analysis Critical Control Points (HACCP) requiring record keeping.
Thus, it is important to provide the agricultural industry with methods to quantify and correlate
sensory and instrumental evaluations of foods.
Machine vision imitates human visual perception by using a camera and a computer with
software capable to generate precise, consistent, and cost-effective color measurement. The
comparison and correlation of instrumental and visual color analysis has been performed in
many uniformly colored agricultural products such as meat, bakery and seafood. Generally,
there is a close relationship between sensory and instrumental color analysis of homogenous
foods. However, comparison and correlation of non-homogeneous color measurements in foods
is more challenging and has not been thoroughly studied.
14
Machine vision was used to quantify the degree of color uniformity of mangos and
nectarines using the number of color blocks and color primitives. The use of color primitives
provided a more accurate method to measure color uniformity of mangos and nectarines. Three
reference color bars (8, 12 and 16 colors) were created from color analysis of the fruits. A
sensory panel (n=80) visually evaluated mangos and nectarines in two presentations: screen
images captured by machine vision and fruits placed in trays. Panelists attempted to quantify
color by selecting (2, 4 or 6 colors) from the reference color bars and compare the colors in the
reference bars with those of the fruit surfaces. There were a total of 9 sessions at different days
using different panelists.
Sensory and machine vision evaluations were compared using the absolute ΔE value. ΔE
measures total color change by accounting for combined changes in L*a*b values. The concept
of the best possible ΔE or best performance under given circumstances was also evaluated. It
was apparent that the number of reference colors and color selections had an impact on the error
made by panelists. More color selections reduced the ΔE values of the visual evaluations.
Statistical analysis described significant differences between the number of reference colors, the
number of selections, presentation, and the interaction between the reference colors and the
selections. The 8 and 16 reference colors bar provided less error compared to the 12 reference
colors bar, quantified by both ΔE for both mangos and nectarines. The 12 reference colors bar
gave the most error. Two color selections provided the highest mean values. The screen images
in general had lower mean values than the fruit trays.
This study provided a better understanding of the way panelists perceive non-uniform
colors. It also suggested a new formulation of consumer panel studies involving non-uniform
visual attributes of foods.
15
CHAPTER 1 INTRODUCTION
Today’s consumers have increased expectations for the quality of food they purchase. In
this competitive market there is no second chance to make a first impression. An important first
impression is the color and appearance of food. How do consumers perceive color? Humans
have difficulty in quantifying color, but are good at comparing it with a reference color.
Therefore, reference colors are used in many instances, e.g. color of a potato chip, salmon color,
egg yolk color, etc. In all these examples, the color of the food is relatively uniform. There are a
limited number of studies that correlate the uniform color of foods measured by instruments, and
by sensory panels. However, many foods have non-uniform colors, e.g. mangos, nectarines, etc.
How can we accurately measure the color in this case? Many instruments measure the average
color, but this causes loss of color information in the case of non-uniform foods. Machine vision
technology eliminates this problem by measuring all the colors at the surface of a non-uniform
food. Another difficulty is how to measure the non-uniformity of color. In this study, methods
were developed and used to quantify the non-uniformity of color with the use of machine vision
technology.
Once the non-uniformity of color is determined, how will this affect how consumers
perceive the color of non-uniform foods? Intuitively, we expect that the more non-uniform the
color, the more difficult it will be for consumers to describe or quantify it. In a preliminary
study, we found that for rabbit meat, the more non-uniform the color, the more error consumers
made in correctly quantifying it (Balaban and others, 2007).
In this study, we asked the following questions:
• Can the image of a food material, taken with a good digital camera, and under controlled conditions, be substituted for the real food, for the purposes of evaluating visual and color attributes? If this is possible, then geographical and temporal restrictions in evaluating visual attributes will be eliminated. The image of a food can then be sent anywhere in the
16
world to be evaluated. Food images from different times can be compared without concern for decay. Also, the image of the food, as an accurate representation of it, can be used for record keeping.
• If reference colors are to be used in evaluating the non-uniform color of foods, how many reference colors should be presented to the sensory panelists? How will the number of reference colors affect the error that the consumer makes in quantifying the color? The answer to this question would allow optimization of the number of reference colors to use.
• From a number of reference colors, how many colors should a panelist select? Too few color choices may not allow a good representation of the actual color. On the other hand, too many colors may confuse the panelist, and may allow large errors in the quantification of real colors. The answer to this question will allow the fine-tuning of the way panelists are asked to evaluate non-uniform colors. The quantification of color is becoming increasingly important due to an emphasis on
international trade, and implementation of Hazard Analysis Critical Control Points (HACCP)
requiring record keeping. Thus, it is important to provide the agricultural industry with methods
to quantify and correlate sensory and instrumental evaluations of foods.
The overall impact of this study will be a better understanding of the way panelists
perceive non-uniform colors. This will result in a better formulation of consumer panel studies
involving visual attributes of foods.
17
CHAPTER 2 LITERATURE REVIEW
Color of Foods and Agricultural Materials
Color is an important quality attribute for almost every agricultural product (Delwiche,
1987). Consumers may perceive color as an indicator of freshness and wholesomeness, and
color may affect their final decision to accept/reject food. For the meat industry, muscle color is
the primary characteristic consumers consider when evaluating the quality and acceptability of
meats (Cornforth, 1994). The discoloration of retail beef accounts for $1 billion in price
discounts annually (Mancini and Hunt, 2005). Color determines the degree of ripeness of many
vegetables and fruits (Polder and others, 2000). Different grains and their varieties are
commonly characterized according to kernel color and quality defects such as grass-green, bin-
burnt, and fungal-damaged (Lou and others, 1999).
Color measurement of food and agricultural materials can be performed subjectively by
sensory panels (Chizzolini and others, 1993). Color can also be measured by instrumental
methods (Balaban and Odabasi, 2006). The quantification of color is becoming increasingly
important due to an emphasis on international trade, and implementation of Hazard Analysis
Critical Control Points (HACCP) requiring record keeping.
Instrumental Color Measurement in Agricultural Food Products
The agricultural industry uses mostly high cost, labor intensive methods to assure control
of color quality parameters. One possibility to reduce cost is to use instrumental methods to
measure color to emulate human visual perception (Zhu and Brewer, 1999). Instruments are
cost-effective, repeatable and objective in measuring color. Instruments such as colorimeters are
commonly used to measure color in the agricultural industry. Colorimeters provide users with
fast and simple “averaged” color measurements.
18
The accuracy of the instrument is assured by calibrating with standard color tiles before
measurement. The color reading is obtained by providing a controlled illuminant or standard
light source. Common standard light sources are: A=tungsten lamp, B= near sunlight, C= near
daylight, D= daylight. Colorimeters have illuminants C or D65 with color temperatures of 6774°
K or 6504° K (Oliveira and Balaban, 2006). However, it is known that other methods provide
more precise color measurements (Coles and others, 1993). Colorimeters may not measure the
observed color if the product has non-uniform colors, because all colors in their view area are
averaged. If the agricultural product is too small, or too big, or has non-uniform surfaces, then
sampling location for color measurement becomes critical. Also, careful consideration is
necessary if data are compared between industrial plants, since variations between instruments
may occur (Brewer and others, 2001).
Spectrophotometers are also used in agriculture to measure color. The working method of
these instruments is based on the generation of a spectral curve representing the transmittance or
reflectance of light from the surface of the product. This is immediately compared with the
reflectance of a reference standard. The values may be converted to different color space values.
The agricultural industry requires a better method of color measurement. In the 1960s the
use of a camera with a computer and software capable of image processing became an option for
color measurement (Brosnan and Sun 2004). The system was called computer vision or machine
vision. The capabilities of this instrument were precise, accurate and fast color measurement of
agricultural products.
Computer Vision or Machine Vision System
The computer vision or machine vision (MV) systems started in the early 1960s. Since
then, the use of machine vision in the agricultural industry has grown. Machine vision is used
for its generation of precise data, consistency, and cost effective color measurement.
19
This instrument aims to emulate human visual perception by using a camera and a
computer with software capable of performing predefined visual tasks (Brosnan and Sun 2004).
Images are captured in digital form by a charge coupled device (CCD) camera. CCD cameras
can convert light into electrical charges and create high-quality, low-noise images with pixels.
They have excellent light sensitivity; they are free of geometric distortion and highly linear in
their response to light (Du and Sun 2004). The computer software then performs image
processing, which is the study of representation and manipulation of pictorial information
(Martin and Tosunoglu 2000). The pictorial information is converted to three-dimensional color
space of red (R) green (G) and blue (B) values. Further analysis provides color results.
Search for cost-reduction and increased efficiency in quality inspection has made the
agricultural industry look for techniques and instruments that provide more complex and
accurate as well as fast and objective determination of quality parameters in online inspection.
Machine vision has shown to be a useful method in this area (Blasco and others, 2003; Lee and
others 2004).
Machine vision has several other advantages over other color measurement instruments:
• Images are composed of the entire view area making the analysis more representative
• The data provided from images can be converted to different color measurement systems (O’Sullivan and others 2003) and processed beyond the capabilities of colorimeters
• Non-uniform surfaces and colors can be handled easily
The agricultural industry uses image processing and MV to classify, sort and grade
agricultural produce in diverse areas such as bakery, meats and fish, vegetables, fruits, grains,
prepared consumer foods and even food container inspection. The food industry ranks among
the top ten industries to use image processing techniques (Gunasekaran, 1996).
20
Bakery Products
Bakery products are influenced by their external as well as internal appearance.
Consumer’s judgment on their appearance dictates purchasing decision and marketability, and it
is essential to meet and exceed their expectations of quality of bakery products. At the same
time it is essential to reduce cost. A MV system was used to classify defective bread loaves by
height and top slope (Scott, 1994). Cookies were studied to estimate the fraction of top surface
area covered with chocolate chip, and other physical features such as size, shape and color of
baked dough (Davidson and others, 2001). MV was capable of providing automated inspection
and could separate light from dark muffin samples (Abdullah and others, 2000).
Red Meat and Seafood
In 2006, the retail value of U.S. beef industry was $71 billion (USDA, 2006 a). More than
12 billion kilograms of beef were consumed in the U.S. in the same year, and, the beef industry
represented 4.4% of U.S. total production exports. In the U.S. nearly 15% of retail beef is
discounted due to surface discoloration, which corresponds to annual revenue losses of $1 billion
(Mancini and Hunt, 2005). The USDA beef carcass grading system consists of two parts: quality
grade and yield grade. Quality grade is evaluated by trained individuals. MV has been
recognized as an objective alternative to assessment of meat quality from fresh-meat
characteristics (Tan, 2004). Recent studies indicate MV has great capability for classification
and grading of beef muscle type, breed, age and tenderness (Basset and others, 2000; Hatem and
others, 2003; Li and others, 1997).
The purpose of grading meats is to standardize the characteristics valuable to the consumer and
those that facilitate marketing and merchandising (Hatem and others, 2003). Beef rib eye steaks
were effectively graded for quality attributes such as color and marbling scores determined by
21
USDA using image processing (Gerrard and others, 1996). The results reported that MV
predicted color with an accuracy level of R²=0.86 and marbling with R²=0.84.
The pork industry has also applied MV to its processes. Pork loins were graded according
to color. Researchers used image processing with statistical and neural network models to
predict color scores of 44 pork loins (Lu and others, 2000). The scores were then compared with
trained sensory panel scores. The scores were based on visual perception ranging from 1 to 5.
Prediction error was the difference between instrumental and sensory scores. An error of 0.6 or
lower was considered not significant. Image processing and neural network models were able to
predict 93.2% of the samples with error lower than 0.6. Statistical regressions were able to
predict 84.1% of the samples with error lower than 0.6. Another study reported 90% agreement
between a MV color score and a sensory panel using 200 pork loin chops (Tan and others, 2000).
Tedious human inspection and costs are part of the grading practices in the poultry
industry. MV was used to separate defective (tumors, bruises, and torn skin and torn meat)
poultry carcasses from normal carcasses (Park and others, 1996).
In 2006, freshwater and marine fishing produced 60 million tons for human consumption
(FAO 2006). Americans consumed an average of 2.2 billion kg of seafood in 2006 (NOAA
2007). Fish represent one of the main sources of protein used in developing countries (Louka
and others, 2004).
Seafood inspection involves costly human involvement. MV was used to capture, identify
and differentiate images of three different varieties of fish: carp (Cyprinus carpio), St. Peter’s
fish (Oreochromis sp.) and grey mullet (Mugil cephalus) (Zion and others, 1999). This study
also concluded that fish mass, an important quality parameter in marketing, could be predicted
from image area with the use of image processing. Other parameters important to market these
22
three types of fish were acquired. Fish species have also been sorted according to shape, length
and orientation in a processing line (Strachan, 1993). Image analysis was used to differentiate
between stocks of Haddock (Melanogrammus aeglefinus) (Strachan and Kell, 1995). Dimension
reduction derived from principal component analysis and canonical correlations was used. The
reports showed 71.7% correct sorting accuracy for shape and 90.9% and 95.6% for both stocks in
color differences. Flesh quality is important for successful development of fish farming and fish
processing (Marty-Mahe and others, 2004). Objective criteria to predict flesh redness from the
spawning coloration of fall chum salmon has been performed with image processing (Hatano and
others, 1989). Skin color development is an important quality parameter for live goldfish
(Carassius auratus), an ornamental fish of high commercial value (Chapman and others, 1997).
Objective measurement and quantification of the color of live goldfish (Carassius auratus)
raised in well water was acquired by a machine vision system (Wallat and others, 2002). The
color of dried cod fillets may go from yellow to orange, depending on the drying method used.
Image processing has been used to compare drying methods in cod fillets (Louka and others,
2004). The fillets were subject to three drying methods: hot air drying, vacuum drying, and
freeze drying. Image processing compared the three techniques to controlled instantaneous
discharge (DIC) and dehydration by successive discharge (DDS), two new techniques of drying
cod fillets. The highest whiteness value found was quantified in freeze-drying and the lowest in
air drying. Analysis of variance was used to find differences between procedures. Vacuum
drying and DIC did not have significant differences.
Catfish ranks as the fourth most popular seafood consumed in the U.S. Fresh farm-raised
catfish (Ictalurus punctatus) quality relies primarily on human inspection. MV was used to
evaluate color changes over storage time for fresh farm-raised catfish (Korel and others, 2001).
23
Vegetables
Vegetables are greatly affected by quality factors such as size, shape, color, blemishes, and
diseases. Image processing resulted in more precise color measurement for potato crisp color
(Coles and others, 1993). The potato industry used a MV system and online inspection to grade
potatoes by shape (Tao and others, 1995). This study reported 89% agreement between the
instrument and human perception. The accuracy in grading potatoes was 90% by using hue,
saturation and intensity color system.
Discoloration of the mushroom cap reduced product quality, with less market value
(Brosnan and Sun, 2004). In order to maximize quality parameters, a MV system was used to
inspect and grade mushrooms based on color, stem cut, shape and cap veil opening (Hienemann
and others, 1994). MV resulted in a 20% classification error compared to two human inspectors.
The surface color of tomatoes was analyzed using a MV system classifying differences in
ripeness stages (Polder and others, 2000). Image processing from MV was used to recognize and
estimate cabbage size for a selective harvester (Hayashi and others, 1998). Surface defects,
curvature and brakes of carrots are quality parameters that influence the product’s value. MV
was used to classify standard and defective carrots (Howarth and others, 1992).
Fruits
In 2005, the U.S. fruit consumption averaged 128 kg per person (fresh-weight equivalent)
(USDA, 2006b), with bananas being the most consumed fresh fruit. Apples were the second
favorite fresh fruit. A MV system was used to evaluate the color to determine the ability of
oxalic acid to inhibit browning in banana and apple slices (Yoruk and others 2004). Golden
delicious apples were evaluated for quality parameters such as bruises, scabs, fungi or wounds
with the use of a MV system (Leemans and others 1998). The results suggested that image
24
processing with different algorithms were able to detect bruises, scabs, fungi and wounds in
golden delicious apples.
In 2002, the U.S. was the world’s largest importer of mangos (Perez and Pollack, 2002).
In this same year, Mexico shipped over 90% of its exports to the U.S. The increased
consumption of mangos is related to the increased population of Latino and Asian groups.
Consumers seek mangos without external damage, with stable weight, color and consistency, at a
reasonable price (Zuñiga-Arias and Ruben 2007). However, grading mangos for export involves
hand labor and subjectivity. A MV system equipped with cameras to obtain single and multiple
view image angles was used to evaluate physical parameters like: projected area, length, width,
thickness, volume, and surface area with 96.47% accuracy (Chalidabhongse and others, 2006;
Yimyam and others, 2005).
Prepared Consumer Foods
The evaluation of cheese functional properties such as different cooking conditions, size of
samples and shred dimensions are important aspects for the marketability of pizza. Topping
types, percentage and distributions influence the appearance and the different varieties of pizza.
Pizza image acquisition is very complex due to the non-homogenous colors, shapes, overlapping,
shadows, and light reflection. Methods have been developed to quantify the color distribution
and topping exposure in pizza (Sun and Du, 2004).
Food Container Inspection
MV and image processing are used to determine shape, and check for foreign matter,
threads of bottles, sidewalls and base defects, fill levels, correct closure and label position of
food containers. MV has also been used to check for wrinkles, dents and other damages to
aluminum cans that cause leakage of contents (Seida and Frenke, 1995).
25
Grains
Nigeria is the world’s leading importer of wheat (USDA 2006c). Ninety percent of the
imported wheat is supplied by the U.S. Competitive prices and product quality has lead the U.S.
increase the wheat market in Nigeria. The variety, environmental effects and class make the
classification of wheat a very complex practice even for experienced inspectors. MV systems
and image processing have been used widely in wheat (Uthu, 2000; Majumdar and Jayas, 2000;
b; Nair and others, 1997). A MV system and crush force features were used to differentiate hard
and soft wheat varieties (Zayas and others, 1996). The correct differentiation rate was 94% for
the varieties tested. Corn kernels were analyzed with MV for whiteness, mechanical and mold
damage (Liu and Paulsen, 1997). Rice has also been studied using MV and image processing.
The appearance characteristics of brown rice such as kernel shape, color, and defects were
determined using a MV system (Wan and others, 2000). An online automatic inspection system
was able to recognize cracked, chalky, broken, immature, and damaged brown rice kernels.
Other Applications
A MV system was used for online inspection of dry sugar granules and powders to
determine particle size for process control and quality improvement (Strickland, 2000). Image
processing and MV were used to detect dirt on brown eggs with stains, dark feces, white uric
acid stains, blood stains and stains caused by egg yolk (Mertens and others, 2005). The results
reported 91% overall accuracy of image processing to detect dirty eggs.
Research efforts were made to provide efficient image-based techniques to monitor
distribution and migration of fish (Nery and others, 2005). Image processing was used to
classify nine species of fish based on adipose fin, anal fin, caudal fin, head and body shape, size
and length/depth ratio of body (Lee and others, 2003). This method provided an alternative to
subjective monitoring of numbers, size and species at specific fish passages during migration.
26
Visual Texture Analysis
Visual texture is defined as how varied or patchy the color of a surface looks (Balaban,
2007). MV systems have been used in determining color, size and shape of agricultural produce.
Texture analysis with MV has great potential due to the powerful discriminating ability and
pattern recognition of this technique.
Texture information may be used to enhance the accuracy of color measurements
(Mäenpää, 2003). Texture is characterized by the relationship of the intensities of neighboring
pixels (Palm, 2004). Visual texture discriminates different patterns of images by extracting the
dependency of intensity between pixels and their neighboring pixels (Kartikeyan and Sarkar,
1991). In other words, texture is the repetition of a basic pattern. The patterns can be the result
of physical surface properties such as roughness or oriented strands, even the reflectance
differences given by a color on a surface (Tuceryan and Jain, 1998).
Visual texture analysis is divided into four main areas: statistical texture, structural texture,
model-based texture, and transform-based texture. Statistical texture describes mainly regions in
an image through high-order moments of their grayscale histograms (Bharati and others, 2004).
Structural texture is described as a composition of elements regulated by rules in images.
Model-based texture generates an empirical model of each pixel in the image based on a
weighted average of the pixel intensities in its neighborhood. Transform-order texture converts
the image into a new form using spatial frequency properties of the pixel under consideration of
its intensity variations.
Image analysis literature describes many ways to quantify texture (Bertrand and others,
1992; Mao and Jain, 1992; Reed and Du Buf, 1993; Tuceryan and Jain, 1998). A new method
used to quantify non-uniform colors is that of color primitives and color change index (Balaban,
27
2007). The methodology used to quantify color non-uniformity should be independent of
rotation, variation in size and shape (Zheng, 2006).
Visual Texture Applications in Agriculture
Image texture analysis has been used in grading and inspection for quality and safety of
agricultural products. A MV system was used with image processing and texture analysis to
quantify changes in color, shape and image texture of apple slices (Fernandez and others, 2005).
A method for texture analysis was developed to quantify non-homogeneity of color of mangos,
apples and rabbit meat using color primitives and color change index, were a color primitive was
defined as a continuous area of an image with similar light intensity (Balaban and others, 2007).
Texture analysis was used to identify the changes in textural appearance in experimental
breads caused by variations of surfactants added to flour (Bertrand and others, 1992). Iyokan
orange fruits (Miyauchi Iyokan) were used to predict sugar content of oranges (Kondo and others,
2000). Image processing and texture analysis were entered to a neural network. MV system
along with neural networks recognized relatively sweet fruit from reddish color, low height,
medium size and glossy surface. Several studies on meat tenderness characteristics
(Chandraratne and others, 2006) and classification of genotypic origins of bovine meat (Basset
and others, 2000) have been successful. Texture analysis evaluated the microstructure of food
surfaces such as potatoes, bananas, pumpkins, carrots, bread crust, potato chips and chocolates
(Quevedo and others, 2002). Texture features have demonstrated to be effective discriminating
models for classifying wholesome and unwholesome chicken carcasses (Park and others, 2002).
Correlation between Image and Visual Color Analysis
The majority of studies regarding color comparison between sensory and instrumental
measures in foods have been developed in the meat area. Research was performed to compare
28
and correlate homogeneous color measurements of pork, beef, and chicken using instrumental
and visual color analysis (Denoyelle and Berny, 1999; Lu and others, 2000; Sandusky and Heath,
1998; Zhu and Brewer, 1999). Meat and poultry were used to correlate instrumental and visual
color evaluation. A range of meat redness was studied by mixing ground poultry breast and
ground beef. High correlations between visual redness and instrumental redness were found
(Zhu and Brewer, 1999).
The comparison and correlation of instrumental and visual color analysis has also been
studied in bakery, seafood, and in medical fields. Research using cookies for color analysis
showed a strong correlation between sensory and instrumental methods (Kane and others, 2003).
The relationship between sensory and instrumental correlations using raw, baked and smoked
flesh of rainbow trout (Onchoyhychus mykiss) was studied. Close relationship between color
evaluation by sensory analysis and instrumental methods was observed (Skrede and others,
1989). A study on colorimetric assessment of small color differences on translucent dental
porcelain revealed strong correlation between instrumental and visual color analysis (Seghi and
others, 1989). However, comparison and correlation of non-homogeneous color measurements
in foods is more challenging and has not been thoroughly studied.
Preliminary Experiments
A method was developed to quantify the perception of non-homogeneous colors of foods
by sensory taste panels. The average colors of mangos, apples, and rabbit meat were measured
using MV. Differences between the average (real) colors (MV system) and those from the
sensory panel were reported as ΔE values (Balaban, 2007).
A sensory panel composed of 20 panelists performed visual evaluations of rabbit meat captured
images and 60 panelists for that of real fruit and captured sample images. The degree of non-
29
uniformity of sample colors was determined using two methods: color blocks and color
primitives.
A color reference bar was developed for the panelists to select colors that represented those of
the samples. Panelists selected 3 colors from these reference colors, and estimated their
percentages. The “red mango” had more color blocks, and visually represented more non-
uniform colors. In the case of rabbit meat samples, there was no apparent advantage of using the
color block scheme. Clearly, a different method to quantify non-uniformity needed to be
developed for these samples. The rabbit samples had colors ranging from white to red, with
many shades in between. The lack of any other hue value may have contributed to the inability
of the color block scheme to quantify non-uniformity.
The more non-uniform samples were more difficult to evaluate, thus, the ΔE error was
higher. The non-uniformity of the samples caused more difficulty in the panelists’ matching
ability with the reference color scale, and caused higher errors.
Males (33) and females (27) were compared regarding ΔE values. The mean ΔE for males
and females was 10.58 and 10.18, respectively, with a p-value= 0.52. In this study gender did
not significantly affect ΔE. A higher number of panelists may or may not affect this outcome.
This preliminary research suggested a criteria and parameters to quantify the error panelists
made when subject to visual appraisal of non-homogenous colors in foods (Balaban, 2007;
Balaban and others, 2007). However, the number of colors that panelists selected from a
reference color bar was limited to 3 choices. More studies are needed to study the effect of the
number of colors in the reference scale, and the number of colors to choose.
The food industry could benefit from a better understanding of precise, repeatable and
accurate color measurements of foods with non-uniform surfaces and/or colors. The quantitative
30
measurement of color attributes of agricultural materials is important in quantifying quality,
maturity, defects, and various other color-dependent properties. Global market expansion and
implementations of Hazard Analysis Critical Control Points (HACCP) require record keeping.
The difference of screen captured image and real sample and its effect on human perception of
sensory evaluations has not been studied thoroughly. A properly taken image of a food sample
can be a good representation of the food itself. This may provide a usable and more flexible tool
in the analysis of visual attributes.
Objectives of the Study
The objectives of this study were:
i. To measure differences in color evaluation between sensory panel and MV system, for non-uniformly colored fruits and their images.
ii. To develop a quantitative measure of the degree of non-uniformity of color, and to
evaluate the effect of degree of non-uniformity of sample color on the difference in color evaluation.
iii. To evaluate the effect of the number of reference colors, and number of allowed color
selections on the error in color evaluation
31
CHAPTER 3 MATERIALS AND METHODS
Mangos and Nectarines
The fruits used in this study were artificial fruits to avoid color degradation due to
maturation and decay of real fruits. Mangos and nectarines generally have non-uniform colors
and surfaces. The fruits used in this study consisted of red mangos and nectarines with non-
uniform surface colors. The mangos were purchased from Amazing Produce (4470 W. Sunset
Boulevard Suite 106 Los Angeles, CA 90027) and the nectarines made of compressed polyfoam
from Zimmerman Market (254 E Main St Leola, PA 17540) (Figure 3-1). The fruits were placed
on aluminum trays. Adhesive tape was used to keep fruits from moving while images were
captured. There were a total of 10 trays with one mango and one nectarine in each. The mangos
and nectarines shown in Figures A-1 to A10 were first wrapped in grey paper (R= 128, G = 128,
B = 128) to obtain a color neutral background.
Figure 3-1. Example of mango and nectarine on aluminum tray.
32
Image Acquisition
The artificial fruits were placed inside a light box built of white acrylic sheets as shown
in Figure A-11. The light box had top and bottom lighting with 2 fluorescent lights each to
simulate illumination by noonday summer sun (D65 illumination). The door remained closed
while images were captured to assure uniformity of light inside and to minimize the effect of
outside light. Images were captured using a camera (Nikon D200 Digital Camera, Nikon Corp.,
Japan) located inside the chamber mounted to face the bottom of the light box as shown in
Figure A-11. The image acquisition set up is shown in Figure A-12. The Nikon D200 Settings
used are described in Table 3-1. After the images were captured, trays were labeled for booth
and tray numbers for identification purposes.
Table 3-1. Nikon D200 settings. Setting Specification
Device Nikon D200 Lens VR 18-200 mm F 3.5-5.6 G Focal length 36 mm Sensitivity ISO 100 Optimize image Custom High ISO NR Off Exposure mode Manual Metering mode Multi-pattern Shutter speed and aperture 1/3s –F/11 Exposure compensation (in camera) 0 EV Focus mode AF-S Long exposure NR Off Exposure compensation (by capture NX) 0 EV Sharpening Auto Tone compensation Auto Color mode Model Saturation Normal Hue adjustment 0 White balance Direct sunlight
Image Analysis
Each captured image included a “red” color standard with known L*, a*, and b* values
(Certified Reflectance Standard, Labsphere, ID# SCS-RD-020). Captured images of the fruits
33
were analyzed for average color, color blocks, and color-texture profiles using MV software.
The values obtained were compared with the measured L*, a*, and b* values of the red standard.
The difference of the L*, a*, and b* values was used as the correction factor for the whole
image. The images were “cleaned” using an image editing software. Each acquired pixel had
(R), (G) and (B) color intensities. The calibrated images were then used to determine the
average L*, a*, and b* values using every pixel of the fruits with Lens Eye color evaluation
software.
For color block analysis, the program read RGB values from every pixel in the captured
image, and counted that pixel a specified color block. Each pixel’s RGB values were converted
first to tristimulus values XYZ, and then to L*, a*, and b* values.
The color data generated by the software was presented in histogram form. This feature
allowed all colors present on the surface area to be seen more easily. Because all colors present
were too numerous to be considered for the color scale formation, a method was developed to
represent the most significant surface colors.
Experimental Design
For this study, a completely randomized design was used. Because the effects of two or
more factors may affect the outcome, whether or not interaction exists, a factorial experimental
design was implemented.
The independent variables considered were number of reference colors, number of colors
to choose from the reference colors, and the sensory evaluation of screen image or real fruit. The
dependent variable for this study was the ΔE values. The ΔΕ value is the color differences
between sensory and MV measured colors of each sample for each panelist. ΔE measures total
color change by accounting for combined changes in L*a*b values.
34
( ) ( ) ( ) 222 ****** sososo bbaaLLE −+−+−=Δ (3-1)
The subscript 0 refers to the MV read values, and s refers to the panelists’ average
The sensory panel combinations are shown in Table 3-2, and each session was performed
at different days using different panelists.
Table 3-2. Factorial-Level combinations. Number of references Two Selections Four Selections Six Selections8 Session 1 Session 2 Session 3 12 Session 4 Session 5 Session 6 16 Session 7 Session 8 Session 9
Method of Selection of the Reference Color Bars
Reference color bars were added to each image to be presented to the panelists (Figure A-
13). Using all 10 fruit tray images (both mangos and nectarines), sixteen global reference colors
were selected from all the color blocks with more than 1% of the surface area of a sample. The
reference color bars consisted of 8, 12 and 16 reference colors (Figure 3-2). Each color in the
reference color bars had known L*a*b values (Tables A-1, A-2, A-3).
The different color scales and number of colors were designed to test and quantify the
effect of these variables on the ability of panelists to match fruit colors. From our preliminary
study, we expected that it would be harder for panelists to correctly select several colors. On the
other hand, more color selection may enhance the ability to predict closer to the real color.
The sample numbers for tray and fruit images were the same. Also, the tray assigned to
each booth was maintained throughout the nine sensory sessions. However, the presentation of
the images or fruits was randomized.
35
Booth 03
397541
06060101 0303 0404 0505 08080202 0707
Figure 3-2. Example of reference color bar with 8 colors added to fruit images presented to the panelists.
Sensory Evaluations
The sensory panel was composed of college age students from the University of Florida.
Each session consisted of n=80 panelists. There were a total of 9 sessions (1 combination per
session) at different days using different panelists. Panelists evaluated fruits from two sources:
screen image and fruit tray. The presentation order was randomized.
The questionnaires consisted of two separate paper sheets handed to participants at the
stages of the sensory evaluation. Each paper sheet included: evaluation stage (image or tray
sample) date, age and gender of the panelist, booth number, and five instruction steps explaining
how to fill out the questionnaire. Also, each sheet included a table with two columns (Figure A-
14 and A-15). The two columns in each table included spaces to select the sample number of
colors to choose, and percentage of total area per color.
36
Sensory room staff explained to the panelists the reference color bars, and how to match
each color to the surface area of the fruits on the screen or in the tray. Once panelists finished
evaluating, e.g. the screen images, they handed in the questionnaire to staff, who made sure it
was properly filled out. In the second part of the session, using e.g. real fruit trays, staff would
pass the questionnaires and the fruit tray to panelists. They would also explain again the
directions to properly fill the questionnaires. Before the participants could leave the room our
sensory panel staff checked the above and made sure that the panelists followed instructions
properly, filled out evaluation sheets, one for screen image and one for fruit tray. It was critical
for our study that panelists selected the correct number of colors and that percentages added to
100. The questionnaires were checked one more time for selection numbers and percentages and
the data were entered to a spreadsheet to prepare for statistical analysis.
Determination of Color Uniformity of Fruit
Degree of non-uniformity of color or color-texture is a relatively new area in the computer
vision field. The degree of non-uniformity in this study was quantified using two methods: the
number of color blocks, and color primitives (Balaban, 2007).
Average Color:
Individual L*, a*, and b* values of each pixel in an object are read and averaged. For
uniform colored materials this method is satisfactory, but when the colors are non-homogeneous,
the averaging may result in unrealistic colors. For most agricultural materials colors vary
throughout the surface. Therefore an average color is of little use. Also, frequently defects or
ripening stages are detected because they are of different colors.
37
Color Blocks
The machine vision system used in this study captured images having 24 bits of color.
Each acquired pixel had three-dimensional color space RGB color intensities represented by 8
bits in the computer, resulting in 256 possible values for each. The total number of distinct
colors represented by this system (256)3 is too large to apply in reality and a known method was
used to reduce the number of colors in the color space. In this study each color axis was divided
into 16 (16 x 16 x 16 = 4096 color blocks). Any color within a color block was represented by
the center color of that block. The machine vision system then counted the number of pixels that
fell within a color block, and calculated the percentage of that color based on the total view area
of the object. Some color blocks were ignored because their percentage in the total area was too
small. The acceptance threshold for the color blocks was set to 1% of the total area. The
assumption was that the higher the number of color blocks, the more non-homogeneous the color
of the object.
Color Primitives
A color primitive is defined as a continuous area of an image where the “intensity” of any
pixel compared to an “anchor pixel” is within a given threshold value (Balaban, 2007). The
intensity difference is defined as Δ I.
( ) ( ) ( ) 222sososo BBGGRRI −+−+−=Δ (3-2)
Once all the pixels that belong to a primitive with ΔI values less than a given threshold are
found, and no other pixels can be added, then the anchor pixel is changed to an available,
neighboring pixel and the process continues until all the pixels of an object are processed.
The subscript o defines the “base” color, and the subscript s defines a pixel that is tested.
38
The center of gravity of the pixels belonging to a primitive was calculated (Balaban,
2007). Also, an equivalent circle with the same area as that of the primitive was found. The
radius of this circle was defined as:
πareaprimitiveradius = (3-3)
This circle was drawn with its center at the center of gravity of the primitive. The
advantage of the color primitives is that there may be many primitives with the same color, but
they will be counted separately. The more color primitives in an image, the more non-
homogeneous the color of that object. The ΔI values of neighboring color primitives, and the
distance between their equivalent circles can be used to quantitatively calculate the degree of
non-homogeneity of color.
circlesequivalenttwobetweenceDisIcolorofchangeofrate
tanΔ
= (3-4)
The more color primitives there are, the higher the value of the cumulative “color change
index”. Also, the bigger the area of an object the more color primitives, everything else being
equal. Therefore, a “color change index (CCI)” was proposed (Balaban, 2007):
100tan areaobject
neighborsofnumbercirclesequivalentbetweencesdis
primitivesgneighborinallforICCI
∑∑Δ
= (3-5)
MV results reported L*, a*, and b* values. Lens Eye color evaluation software required all
reference bar L*a*b values to be entered. It also required the input of all MV determined L*a*b
values from all 10 trays. The final data input to the program was that of the panelists and their
choices of colors from the reference color bars.
The output of the program reported panelists L*a*b values as well as MV determined L*a*b
values. It finally calculated ΔE values for all nine sensory sessions. Each reference color’s L*,
39
a*, and b* values were weighted by the percentage, and averaged, providing the estimated
average color of a sample. The data was grouped in order to prepare for statistical analysis.
iref
n
i
i LPorLAverageColEstimated ,1
*100∑
=
=∗ (3-6)
iref
n
i
i aPraverageColoEstimatedA ,1
*100∑
=
=∗ (3-7)
iref
n
i
i bPrbverageColoEstimatedA ,1
*100∑
=
=∗ (3-8)
The variable n refers to the number of selections (2, 4, and 6). Pi refers to the percentage of
color i. L*ref,I is equal to the L* value of ith reference color. a*ref,I is equal to the a* value of ith
reference colors, b*ref,I is equal to the b* value of ith reference colors.
Calculation of Best Possible ΔE
There is an inherent error in trying to correctly quantify the non-uniform color of a material
using a finite number of reference colors, and a finite number of selections from these colors.
The best possible values given the different selection of reference colors and choices a panelist
could provide are described as “the best possible ΔE”. A computer program was developed to
take each combination of possible choices from a given number of reference colors, and then try
the percentages of these selected colors from 1 to 100, accept the cases with the sum of all
percentages adding to 100, and find the combination with the minimum ΔE. This value can then
be subtracted from the ΔE values of panelists (absolute ΔE) to form a more accurate error term.
Statistical Analysis
The ΔE for sensory and MV measured colors of each sample for each panelist were
calculated. The results were evaluated for the ten booths and nine sessions. The ΔE values were
analyzed using SAS 9.0 for statistical analysis.
40
ANOVA procedures (Duncan’s multiple range tests) and Mixed Models (Restricted
Maximum Likelihood REML LS means) were used to find significant differences on the effect
of reference color bar, number of selections, and presentation source screen image or fruit tray.
The mixed model provided the flexibility of modeling not only the means of data (as in the
standard linear model) but their variances and co variances and fixed effects as well. The need
for covariance parameters arouse because repeated measurements were taken on the same
experimental unit, and these repeated measurements are correlated or exhibit variability that
changes. The ANOVA can provide incorrect results depending on the design because if analysis
is driven by accounting for degrees of freedoms and tests, p-values, contrasts, least square means
etc. may be taken for granted. The ANOVA used in this experiment computed means of the
dependent variables for the effects mentioned earlier based on Ordinary Least Squares. All main
effects were tested using means for those effects.
In each method, ΔE absolute or ΔE difference were used as model statements or dependent
variables. The program codes and outputs are shown in Tables E-1 and 2.
41
CHAPTER 4 RESULTS & DISCUSSION
MV Color Results of Fruits
MV measured colors of each fruit in each booth are summarized in Tables 4-1 and 4-2.
For both mangos and nectarines it can be seen that L*a*b values and number of color blocks are
similar for all 10 booths. It is also seen that the Color Change Index (CCI), the number of
primitives and the number of neighbors are slightly different for each booth.
Table 4-1. MV color analysis for mangos. Booth L* a* b* # Color Blocks CCI # Primitives # Neighbors 1 51.79 31.91 42.39 31 6.423 616 1190 2 52.92 27.18 45.61 31 9.218 613 1239 3 49.62 37.21 44.4 33 5.588 513 982 4 50.42 25.31 42.91 33 4.904 548 1012 5 48.41 28.24 42.62 22 6.968 593 1127 6 56.39 21.93 46.84 30 7.164 615 1190 7 48.48 39.19 42.4 32 6.762 609 1196 8 46.88 38.92 40.34 20 4.338 538 964 9 46.64 36.3 40.59 33 4.589 536 1016 10 52.22 31.26 43.85 32 11.294 738 1511
Table 4-2. MV color analysis for nectarines. Booth L* a* b* # Color Blocks CCI # Primitives # Neighbors 1 52.57 36.94 43.81 31 17.516 915 1760 2 53.3 35.55 45.31 31 13.721 693 1366 3 53.16 34.67 48.11 32 13.135 581 1196 4 54.61 35.67 41.26 27 13.202 794 1521 5 62.78 26.47 50.46 28 11.326 626 1215 6 60 28.71 49.91 31 14.231 729 1491 7 65.25 24.48 45.09 25 17.759 856 1648 8 53.45 38.59 44.06 30 15.194 756 1488 9 56.51 30.86 43.92 31 11.490 739 1414 10 57.57 29.13 42.24 30 14.768 773 1566
42
The tables A-4, A-5 show a summary of the number of primitives, number of neighbors
and color change index for mangos and nectarines used in this study. The color primitives
analysis, e.g. mango and nectarine for booth 1, is shown in Figure A-16.
Non-Uniformity Analysis of Fruits
The degree of non-uniformity was determined using color blocks and number of primitives
(Tables 4-1, 4-2). The numbers of color blocks considered were those colors greater than 1% of
the sample surface area. The average L*a*b values in all booth for mangos and nectarines were
similar to each other. Because of these similarities between booths, the number of color blocks
were also similar between booths. The number of color blocks for the two fruits had a range of
values from 20-33, with higher values being predominant. However, in some instances, e.g.
booth 8 for mangos and booth 7 for nectarines, the number of color blocks was 20 and 25,
respectively (Tables 4-1 and 4-2). For nectarines, the range of color blocks was from 32 to 25,
representing a change of 22%. The CCI values ranged from 17.7 to 11.3, representing a change
of 36%. This suggests that the number of color blocks for mangos and nectarines is not a better
parameter to quantify the non-uniformity of these samples, compared to CCI. This was expected,
since from previous results it was considered that the number of color blocks does not represent
an accurate method to quantify colors of non-homogenous foods with a wide distribution of
hues.
Although the number of color blocks in most booths were close, it can be seen that the
number of primitives, number of neighbors, and CCI change for each booth were different. For
example, the CCI for mango in booth 1 was 6.4 and that for booth 2 was 9.2. For both fruits, the
number of color blocks was 31. The same pattern is seen for booth 1 and booth 6 of the
nectarines as shown in Table 4-2.
43
The correlation between CCI, number of primitives and number of neighbors for both
mangos and nectarines is shown in Figures 4-1, 4-2 and 4-3.
0 5 10 15 20CCI
400
500
600
700
800
900
1000
#Prim
itive
s
Mango Nectarine
Figure 4-1. Correlation between number of primitives and color change index (CCI).
0 5 10 15 20CCI
900
1000
1100
1200
1300
1400
1500
1600
1700
1800
#Nei
ghbo
rs
Mango Nectarine
Figure 4-2. Correlation between number of neighbors and color change index (CCI).
R-square = 0.8471
R-square = 0.7895
44
400 500 600 700 800 900 1000#primitives
900
1000
1100
1200
1300
1400
1500
1600
1700
1800
#Nei
ghbo
rsMango Nectarine
Figure 4-3. Correlation between number of neighbors and number of primitives.
From these figures it can be seen that these three variables are strongly correlated, and
therefore can be used interchangeably. The number of primitives, CCI and the number of
neighbors effectively quantified the non-uniformity of mangos and nectarines in all 10 booths.
In general, nectarines were more non-uniform than mangos. It would not be possible to make
this conclusions using the L*a*b average values or the number of color blocks.
Best Possible ΔE
Tables 4-3 to 4-10 show the best possible ΔE for various combinations of color
references and selections. The cases with 6 selections of colors are not shown since all best ΔE
values were less than 1. It can be seen that especially for 2 or 4 selections, the best ΔE values
can be significantly high. This means that there is an inherent error associated with selection of
2 or 4 reference colors regardless of the number of available reference colors. Therefore, the
“real” error that a panelist makes in estimating the color of a sample is the difference between
R-square = 0.9685
45
the “absolute ΔE” and the “best possible ΔE”. Large “best ΔE” values are not restricted to 2
selections only. This is shown in Table 4-8 with 4 selections out of 8 reference colors in booth 7.
Table 4-3. Best possible selections and minimum ΔE value possible for 8 references and 2 selections for mangos.
Booth Color 1 1% Color 2 2% Min ΔE 1 3 56 5 44 2.923 2 3 46 5 54 1.895 3 3 67 8 33 2.407 4 3 14 4 86 2.588 5 1 42 3 58 1.931 6 4 87 7 13 2.669 7 3 71 8 29 2.191 8 3 70 5 30 0.827 9 3 46 4 54 0.612 10 3 54 5 46 2.800
Table 4-4. Best possible selections and minimum ΔE value possible for 8 references and 2
selections for nectarines. Booth Color 1 1% Color 2 2% Min ΔE
1 3 65 8 35 1.337 2 3 62 8 38 0.907 3 3 59 8 41 1.893 4 3 63 8 37 4.659 5 3 40 8 60 2.647 6 3 45 8 55 1.267 7 3 40 8 60 8.666 8 3 59 7 41 1.247 9 3 55 8 45 4.646 10 3 53 8 47 7.079
Table 4-5. Best possible selections and minimum ΔE value possible for 12 references and 2 selections for mangos.
Booth Color 1 1% Color 2 2% Min ΔE 1 3 56 5 44 2.923 2 3 46 5 54 1.895 3 3 67 8 33 2.407 4 3 14 4 86 2.589 5 1 42 3 58 1.932 6 4 87 7 13 2.669 7 3 71 8 29 2.192 8 3 70 5 30 0.827 9 3 46 4 54 0.613 10 3 54 5 46 2.801
46
Table 4-6. Best possible selections and minimum ΔE value possible for 12 references and 2 selections for nectarines.
Booth Color 1 1% Color 2 2% Min ΔE 1 3 65 8 35 1.338 2 3 62 8 38 0.908 3 3 59 8 41 1.893 4 3 63 8 37 4.660 5 3 40 8 60 2.647 6 3 45 8 55 1.267 7 3 40 8 60 8.666 8 3 59 7 41 1.247 9 3 55 8 45 4.646 10 3 53 8 47 7.079
Table 4-7. Best possible selections and minimum ΔE value possible for 16 references and 2 selections for mangos.
Booth Color 1 1% Color 2 2% Min ΔE 1 3 56 5 44 2.923 2 3 46 5 54 1.895 3 3 67 8 33 2.407 4 3 14 4 86 2.589 5 1 42 3 58 1.932 6 4 87 7 13 2.669 7 3 71 8 29 2.192 8 3 70 5 30 0.827 9 3 46 4 54 0.613 10 3 54 5 46 2.801
Table 4-8. Best possible selections and minimum ΔE value possible for 16 references and 2 selections for nectarines.
Booth Color 1 1% Color 2 2% Min ΔE 1 3 65 8 35 1.338 2 3 62 8 38 0.908 3 3 59 8 41 1.893 4 3 63 8 37 4.660 5 3 40 8 60 2.647 6 3 45 8 55 1.267 7 3 40 8 60 8.666 8 3 59 7 41 1.247 9 3 55 8 45 4.646 10 3 53 8 47 7.079
47
Table 4-9. Best possible selections and minimum ΔE value possible for 8 references and 4 selections for mangos.
Booth Color 1 1% Color 2 2% Color 3 3% Color 4 4% Min ΔE 1 3 44 4 1 6 43 7 12 2.798 2 3 29 4 1 6 52 7 18 1.377 3 3 15 4 1 6 32 7 52 0.177 4 1 1 3 1 4 17 5 81 2.576 5 1 25 3 1 4 53 5 21 1.792 6 3 19 5 1 6 62 7 18 2.990 7 3 30 4 3 6 28 7 39 0.664 8 3 64 4 1 6 29 7 6 0.686 9 3 3 4 54 5 42 6 1 0.587 10 3 30 4 1 6 44 7 25 2.182
Table 4-10. Best possible selections and minimum ΔE value possible for 8 references and 4 selections for nectarines.
Booth Color 1 1% Color 2 2% Color 3 3% Color 4 4% Min ΔE 1 3 1 4 1 6 33 7 65 2.828 2 3 1 5 1 6 36 7 62 2.564 3 5 1 6 41 7 40 8 18 2.125 4 3 17 4 1 6 36 7 46 5.930 5 5 1 6 59 7 36 8 4 8.814 6 5 1 6 54 7 36 8 9 6.851 7 4 1 5 1 6 60 7 38 11.595 8 4 1 5 1 6 30 7 68 3.688 9 3 1 4 1 6 45 7 53 5.452 10 3 18 4 1 6 49 7 32 7.221
Table 4-11. Best possible selections and minimum ΔE value possible for 12 references and 4
selections for mangos. Booth Color 1 1% Color 2 2% Color 3 3% Color 4 4% Min ΔE 1 4 32 5 47 7 2 9 19 0.0211 2 5 75 7 12 8 2 9 11 0.0225 3 4 8 5 43 7 43 10 6 0.0165 4 1 37 3 6 4 47 9 10 2.3482 5 1 34 3 1 4 62 9 3 1.7248 6 5 72 6 5 9 19 10 4 0.0220 7 1 11 3 58 4 11 10 20 0.0288 8 3 54 5 39 9 5 11 2 0.0352 9 1 18 4 75 5 2 9 5 0.4383 10 3 36 6 37 7 13 9 14 0.0314
48
Table 4-12. Best possible selections and minimum ΔE value possible for 12 references and 4 selections for nectarines.
Booth Color 1 1% Color 2 2% Color 3 3% Color 4 4% Min ΔE 1 4 59 7 8 9 10 12 23 0.0396 2 5 43 8 30 9 26 12 1 0.0192 3 1 14 3 38 8 15 10 33 0.0252 4 3 18 5 26 7 18 9 38 0.0527 5 6 9 7 32 9 34 11 25 0.0318 6 2 23 4 11 9 39 11 27 0.0241 7 5 20 8 1 9 70 11 9 0.0453 8 4 34 7 32 9 20 11 14 0.0327 9 2 22 4 8 6 20 9 50 0.0313 10 1 17 3 6 4 27 9 50 0.0616
Sensory Panel Results
Images and trays of fruits were used in visual sensory evaluations. The reference color bars
and color selections provided to panelists had the following combinations:
1. Treatment 1 = 8 reference color and 2 color selections
2. Treatment 2 = 8 reference colors and 4 color selections
3. Treatment 3 = 8 reference colors and 6 color selections
4. Treatment 4 = 12 reference color and 2 color selections
5. Treatment 5 = 12 reference colors and 4 color selections
6. Treatment 6 = 12 reference colors and 6 color selections
7. Treatment 7 = 16 reference colors and 2 color selections
8. Treatment 8 = 16 reference colors and 4 color selections
9. Treatment 9 = 16 reference colors and 6 color selections
The summary performance of panelists evaluating mangos and nectarines in booth 1 are shown
in Tables 4-13 and 4-14, respectively. The summary performance for the rest of the booths is
shown in Tables B-1 to B-10.
49
Table 4-13. Summary performance for panelists evaluating mangos for booth 1. Case Best ΔE Screen ΔE Stdev Tray ΔE Stdev
1* 2.920 11.389 9.556 15.752 6.046 2* 2.790 10.701 5.158 10.615 2.298 3* 0.200 9.488 4.044 10.519 5.414 4* 2.920 21.152 7.643 18.036 8.864 5* 0.200 13.533 7.794 12.844 3.014 6* 0.200 13.872 8.261 12.729 4.741 7* 2.920 15.198 10.286 14.933 4.890 8* 0.200 13.972 9.509 11.335 1.861 9* 0.200 13.907 5.911 6.404 6.404
The number of reference colors and color selections has an impact on the error made by
panelists as reflected in absolute ΔE values (screen ΔE or tray ΔE). In Table 4-13 the best ΔE
that panelists could obtain for treatment 1 (2 selections) was 2.92. The actual average values
provided by panelists evaluating the screen image was 11.34 with a standard deviation of 9.56,
and that for fruit tray was higher at 17.75 and smaller standard deviation of 6.05. More color
selections reduce the ΔE values of the visual evaluation of mangos. This is shown for ΔE values
in case 2 (4 selections) and case 3 (6 selections) in Table 4-13.
Figure 4-4. Comparison of ΔE values for 8, 12, and 16 reference colors, 2 selections.
0 1 2 3 4 5 6 7 8 9 10 11trays
0
10
20
30
40
Del
ta E
8 ref, 212 ref, 216 ref, 2
Mango, abs.DE, real fruit
50
For mangos, it was apparent that panelists had more difficulty with 12 reference colors
and its combinations. Treatment 4 provided panelists with the most challenge of all
combinations reflecting a high ΔE value of 21.15 for the screen image and 18.036 for fruit tray.
The ΔE values for the combinations with 12 references are higher than the rest of the
combinations (Figure 4-4).
Table 4-14. Summary performance for panelists evaluating nectarines for booth 1. Case Best ΔE Screen ΔE Stdev Tray ΔE Stdev
1* 1.34 9.148 4.788 12.970 8.376 2* 2.82 7.410 2.789 11.314 3.515 3* 0.2 7.610 2.724 10.089 2.376 4* 1.34 25.728 14.359 20.020 14.592 5* 0.2 13.545 9.165 11.070 1.256 6* 0.2 8.770 1.822 10.615 4.044 7* 1.34 13.271 11.203 12.658 1.423 8* 0.2 9.950 6.329 12.572 3.761 9* 0.2 14.134 10.592 10.566 4.915
In Table 4-14 the best ΔE that panelists could obtain for treatment 1 was 1.34. The actual
average value provided by panelists for the screen image of a nectarine was 9.15 with a standard
deviation of 4.79, that for fruit tray was 12.97 and smaller standard deviation of 8.38. In general,
the ΔE values of nectarines were higher than those of mangos.
Statistical Analysis
The data for the sensory panel was analyzed for sources of variation such as number of
reference colors, selection of colors and its interactions, presentation: screen image or fruit tray,
and its interactions with references and selections, using two different models: Mixed model
results, and ANOVA analysis of variance. For the statistical analysis the difference in ΔE was
also used as our dependent variable because this provided a more realistic number due to the best
possible outcome by panelists given the combinations of reference colors and selection of colors.
51
The difference in ΔE was the value obtained from subtracting best ΔE from absolute ΔE. Both
models provided similar data. The data was separately analyzed for mangos and nectarines.
Mangos
The analysis of variance showed significant differences between the number of reference
colors, the number of selections, presentation and the interaction between the reference colors
and the number of selections, both for the ΔE absolute and difference in ΔE (p-value=0.0001) as
shown in Tables 4-15 and 4-16. The mixed mode analysis also reported these same significant
differences as shown in Tables E-1 and E-2. The rest of the interactions did not result in
significant differences for reference color and presentation, selections and presentation and
reference colors, selections of colors and presentation.
Table 4-15. ANOVA summary absolute ΔE for mangos. Source DF* ANOVA SS* MEAN Square F Value Pr >F
Ref. colors 2 2100.72 1050.36 22.47 < .0001 Selections 2 4410.37 2205.18 47.18 < .0001 Presentation 1 10.55.36 1055.36 22.58 < .0001 Ref. colors * selections 4 1948.59 487.15 10.42 < .0001 Ref. colors* presentation 2 189.06 94.53 2.02 0.133 Selections*presentation 2 136.41 68.21 1.46 0.233 Ref. colors*selections*presentation
4 65.66 16.41 0.35 0.843
* Ref. refers to reference. DF refers to degrees of freedom. SS refers to the sum of squares.
Table 4-16. ANOVA summary difference ΔE for mangos.
Source DF ANOVA SS MEAN Square F Value Pr >F Ref. colors 2 2602.76 1301.38 27.73 < .0001 Selections 2 1356.82 678.41 14.46 < .0001 Presentation 1 1055.28 1055.28 22.49 < .0001 Ref. colors * selections 4 1769.15 442.29 9.42 < .0001 Ref. colors* presentation 2 189.43 94.53 2.01 0.134 Selections*presentation 2 136.43 68.21 1.45 0.234 Ref. colors*selections*presentation 4 65.66 16.41 0.35 0.844
* Ref. refers to reference. DF refers to degrees of freedom. SS refers to the sum of squares.
52
It is apparent that the error means for color selections, 2 color choices had the highest mean
and was significantly different than the rest as shown in Figure 4-4 and 4-5. Panelists tend to
make more errors when selecting only 2 colors. The error decreases and panelists become more
efficient with more color choices.
The ANOVA Procedure Duncan's Multiple Range Test for ΔE Alpha 0.05 Error Degrees of Freedom 1422 Error Mean Square 46.74127 Number of Means 2 3 Critical Range .8657 .9115 Means with the same letter are not significantly different. Duncan Grouping Mean N Selections A 15.2401 480 2 B 11.9144 480 4 B 11.2349 480 6
Figure 4-4. Absolute ΔE means difference of selections of colors using mangos.
The ANOVA Procedure Duncan's Multiple Range Test for DiffΔE Alpha 0.05 Error Degrees of Freedom 1422 Error Mean Square 46.92962 Number of Means 2 3 Critical Range .8674 .9133 Means with the same letter are not significantly different. Duncan Grouping Mean N Selections A 13.1554 480 2 B 11.1637 480 4 B 11.0349 480 6
Figure 4-5. Difference in ΔE means difference of selections of colors using mangos.
Because of this, the rest of the color selections, 4 and 6 choices had mean values showing no
significant difference between each other for both ΔE absolute and difference in ΔE.
This same pattern was seen using the mixed mode for statistical analysis and is shown in Figures
E-1 and E-2.
53
It was also apparent that the 8 reference colors provided less error both for the ΔE absolute
and difference in ΔE as shown in Figures 4-6 and 4-7. The highest error made by panelists was
with 12 reference colors compared to 8 and 16. However, nectarines reported slightly higher
error values than mangos. This may be due to the higher non-uniformity of nectarines making the
evaluations harder for panelists. The same results were obtained using the mixed mode as shown
in Figures E-3 and E-4.
The ANOVA Procedure Duncan's Multiple Range Test for ΔE Alpha 0.05 Error Degrees of Freedom 1422 Error Mean Square 46.74127 Number of Means 2 3 Critical Range .8657 .9115 Means with the same letter are not significantly different. Duncan Grouping Mean N Reference Colors A 14.3974 480 12 B 12.5116 480 16 C 11.4803 480 8
Figure 4-6. Absolute ΔE means for reference colors for mangos.
The ANOVA Procedure Duncan's Multiple Range Test for DiffΔE Alpha 0.05 Error Degrees of Freedom 1422 Error Mean Square 46.92962 Number of Means 2 3 Critical Range .8674 .9133 Means with the same letter are not significantly different. Reference_ Duncan Grouping Mean N Colors A 13.4795 480 12 B 11.6834 480 16 C 10.1910 480 8
Figure 4-7. Difference in ΔE Means for reference colors for mangos.
The interaction between the number of reference colors and the number of color selections
also reported significant differences. The highest error made by panelists was when evaluating
treatment 4 or 12 reference colors and 2 color choices as shown in Figures 4-8 and 4-9 both for
absolute ΔE and difference in ΔE.
54
The ANOVA Procedure Duncan's Multiple Range Test for ΔE Alpha 0.05 Error Degrees of Freedom 1431 Error Mean Square 47.45812 Number of Means 2 3 4 5 6 7 8 9 Critical Range 1.511 1.591 1.644 1.684 1.715 1.740 1.761 1.779 Means with the same letter are not significantly different. Duncan Grouping Mean N TRT A 19.1266 160 4 B 13.9341 160 7 C B 12.6597 160 1 C B 12.4915 160 5 C D 11.8113 160 9 C D 11.7896 160 8 C D 11.5741 160 6 C D 11.4620 160 2 D 10.3192 160 3
Figure 4-8. Absolute ΔE means for interaction between the number of reference colors and the number of selections.
The lowest error made by panelists was with treatment 3 or 8 reference colors and 6
selections both for absolute ΔE and difference in ΔE. The rest of the treatments were slightly
different however, providing significant differences. It is clear that the more color selections, the
less error made by the panelists. It is possible that up to certain level of reference colors panelists
would perform more efficiently, and above that level it would too complicated for the panelists
to refer to color selections and reference colors. There may be an optimum number of reference
colors.
55
The ANOVA Procedure Duncan's Multiple Range Test for DiffΔE Alpha 0.05 Error Degrees of Freedom 1431 Error Mean Square 47.64525 Number of Means 2 3 4 5 6 7 8 9 Critical Range 1.514 1.594 1.647 1.687 1.718 1.743 1.765 1.783 Means with the same letter are not significantly different. Duncan Grouping Mean N TRT A 17.0419 160 4 B 12.0225 160 5 B 11.8493 160 7 C B 11.6113 160 9 C B 11.5896 160 8 C B D 11.3741 160 6 C B D 10.5750 160 1 C D 10.1192 160 3 D 9.8789 160 2
Figure 4-9. Difference in ΔE means for interaction between the number of reference colors and the number of selections.
The presentation (screen image vs. fruit tray) was also significantly different (p-
value=0.0001). The fruit tray had mean values higher than the screen image both for absolute
ΔE and difference in ΔE as shown in Figures 4-10 and 4-11. These same results were obtained
using the mixed mode as shown in Figures E-5 and E-6.
The ANOVA Procedure Duncan's Multiple Range Test for ΔE Alpha 0.05 Error Degrees of Freedom 1422 Error Mean Square 46.74127 Number of Means 2 Critical Range .7068 Means with the same letter are not significantly different. Duncan Grouping Mean N Source A 13.6525 720 F B 11.9404 720 S
Figure 4-10. Absolute ΔE means for presentation for mangos.
56
The ANOVA Procedure Duncan's Multiple Range Test for DiffΔE Alpha 0.05 Error Degrees of Freedom 1422 Error Mean Square 46.92962 Number of Means 2 Critical Range .7083 Means with the same letter are not significantly different. Duncan Grouping Mean N Source A 12.6407 720 F B 10.9286 720 S
Figure 4-11. Difference ΔE means for presentation for mangos.
Nectarines
The analysis of variance resulted in significant differences between reference colors, the
number of selections, presentation and the interaction between the reference colors and the
selection of colors for both the ΔE absolute and the difference in ΔE (p-value = 0.0001) as shown
in Tables 4-17 and 4-18.
Table 4-17. ANOVA summary absolute ΔE for nectarines. Source DF ANOVA SS MEAN Square F Value Pr >F
Ref. colors 2 2827.38 1413.69 30.00 < .0001 Selections 2 7657.60 3828.80 81.25 < .0001 Presentation 1 761.75 761.74 16.16 < .0001 Ref. colors * selections 4 3850.96 962.74 20.43 < .0001 Ref. colors* presentation 2 154.12 77.06 1.64 0.195 Selections*presentation 2 122.78 61.39 1.30 0.272 Ref. colors*selections*presentation
4 238.36 59.59 1.26 0.282
* Ref. refers to reference. DF refers to degrees of freedom. SS refers to the sum of squares.
Similar to mangos, the interactions for reference color and presentation, selections and
presentation and reference colors, selections of colors and presentation did not result in
significant differences. These same results were obtained using the mixed mode as shown in
Tables E-3 and E-4.
57
The rest of the interactions did not result in significant differences for reference color and
presentation, selections and presentation and reference colors, selections of colors and
presentation.
Table 4-18. ANOVA summary difference in ΔE for nectarines. Source DF ANOVA SS MEAN Square F Value Pr >F
Ref. colors 2 6968.87 3484.44 75.76 < .0001 Selections 2 1903.92 951.96 20.70 < .0001 Presentation 1 761.82 761.82 16.56 < .0001 Ref. colors * selections 4 4565.35 1141.34 24.82 < .0001 Ref. colors* presentation 2 154.10 77.05 1.68 0.188 Selections*presentation 2 122.79 61.39 1.33 0.264 Ref. colors*selections*presentation 4 238.38 59.59 1.30 0.269
* Ref. refers to reference. DF refers to degrees of freedom. SS refers to the sum of squares.
The same pattern seen previously with mangos were 8 reference colors reported the lowest
value and 12 reference colors the highest value as shown in Figures 4-12 and 4-13 both for the
absolute ΔE and the difference in ΔE. Similar results were obtained using the mixed mode as
shown in Figures E-9 and E-10.
The ANOVA Procedure Duncan's Multiple Range Test for ΔE Alpha 0.05 Error Degrees of Freedom 1422 Error Mean Square 47.12485 Number of Means 2 3 Critical Range .8692 .9152 Means with the same letter are not significantly different. Reference_ Duncan Grouping Mean N Colors A 14.9077 480 12 B 13.0530 480 16 C 11.4792 480 8
Figure 4-12. Absolute ΔE means for reference colors for nectarines.
58
The ANOVA Procedure Duncan's Multiple Range Test for DiffΔE Alpha 0.05 Error Degrees of Freedom 1422 Error Mean Square 45.99134 Number of Means 2 3 Critical Range .8587 .9041 Means with the same letter are not significantly different. Reference_ Duncan Grouping Mean N Colors A 13.6839 480 12 B 11.7746 480 16
C 8.3653 480 8
Figure 4-13. Difference ΔE means for reference colors for nectarines.
The number of selections was also significantly different with (p-value= 0.0001). However,
when looking at the means for color selections, 2 color choices had the highest mean of the rest
of the color selections, 4 and 6 as shown in Figure 4-14 and there were no significant difference
between 4 and 6 color selections of Difference in ΔE as shown in Figure 4-15, the same case as
the mangos. Similar results were obtained with the mixed mode procedures as seen in Figures
E-7 and E-8.
The ANOVA Procedure Duncan's Multiple Range Test for ΔE Alpha 0.05 Error Degrees of Freedom 1422 Error Mean Square 47.12485 Number of Means 2 3 Critical Range .8692 .9152 Means with the same letter are not significantly different. Duncan Grouping Mean N Selections A 16.3154 480 2 B 12.2299 480 4 C 10.8946 480 6
Figure 4-14. Absolute ΔE means for selection of colors for nectarine.
59
The ANOVA Procedure Duncan's Multiple Range Test for DiffΔE Alpha 0.05 Error Degrees of Freedom 1422 Error Mean Square 45.99134 Number of Means 2 3 Critical Range .8587 .9041 Means with the same letter are not significantly different. Duncan Grouping Mean N Selections A 12.8803 480 2 B 10.6946 480 6 B 10.2490 480 4
Figure 4-15. Difference ΔE means for selections of colors for nectarines.
The interaction between reference colors and number of selection of colors also resulted in
significant differences with (p-value = 0.0001).
The ANOVA Procedure Duncan's Multiple Range Test for ΔE Alpha 0.05 Error Degrees of Freedom 1431 Error Mean Square 47.72086 Number of Means 2 3 4 5 6 7 8 9 Critical Range 1.515 1.595 1.649 1.688 1.719 1.745 1.766 1.784 Means with the same letter are not significantly different. Duncan Grouping Mean N TRT A 21.2744 160 4 B 14.9965 160 7 C 12.8476 160 5 C 12.6753 160 1 C 12.4578 160 8 D C 11.7046 160 9 D C 11.3842 160 2 D 10.6010 160 6 D 10.3782 160 3
Figure 4-16. Difference ΔE means for selections of colors for nectarines.
For both the absolute ΔE and difference in ΔE treatment 4 or 12 reference colors and 2 color
choices had the highest value as shown in Figures 4-16 and 4-17. Similar to mangos, panelists
had the most difficulty in matching 12 reference colors and 2 color selections. Panelist also had
difficulty in evaluating treatment 7 or 16 reference colors and 2 color selections reported as the
second highest mean error for absolute ΔE. Panelists performed best and reported the lowest
error value for difference in ΔE with treatment 2 or 8 reference colors and 4 color selections as
shown in Figure 4-16.
60
The ANOVA Procedure Duncan's Multiple Range Test for DiffΔE Alpha 0.05 Error Degrees of Freedom 1431 Error Mean Square 46.59453 Number of Means 2 3 4 5 6 7 8 9 Critical Range 1.497 1.576 1.629 1.668 1.699 1.724 1.745 1.763 Means with the same letter are not significantly different. Duncan Grouping Mean N TRT A 17.8393 160 4 B 12.8114 160 5 B 12.2578 160 8 C B 11.5613 160 7 C B 11.5046 160 9 C D 10.4010 160 6 C D 10.1782 160 3 D 9.2402 160 1 E 5.6777 160 2
Figure 4-17. Difference ΔE means for selections of colors for nectarines.
The presentation (screen image vs. fruit tray) was also significantly different (p-
value=0.0001). The fruit tray had mean values higher than the screen image as shown in Figures
4-18 and 4-19 for both ΔE and Difference in ΔE. Similar results were obtained using the mixed
mode as shown in Figures E-11 and E-12.
The ANOVA Procedure Duncan's Multiple Range Test for Delta_E Alpha 0.05 Error Degrees of Freedom 1422 Error Mean Square 47.12485 Number of Means 2 Critical Range .7097 Means with the same letter are not significantly different. Duncan Grouping Mean N Source A 13.8739 720 F B 12.4193 720 S
Figure 4-18. Absolute ΔE means for presentation for nectarines.
61
The ANOVA Procedure Duncan's Multiple Range Test for DiffDE Alpha 0.05 Error Degrees of Freedom 1422 Error Mean Square 45.99134 Number of Means 2 Critical Range .7011 Means with the same letter are not significantly different. Duncan Grouping Mean N Source A 12.0020 720 F B 10.5473 720 S
Figure 4-19. Difference ΔE means for presentation for nectarines.
ΔE vs. CCI
The number of primitives, number of neighbors, and CCI measure the degree of non-
uniformity of color. ΔE gave the difference between MV and panelists color appraisal. The
correlation between ΔE and non-uniformity (CCI) is shown in Figures C-1 to C-12. The ΔE for
nectarines obtained from screen images and combinations of references of colors and color
selection did not correlate well with the CCI values as shown in Figures C-1 to C-3. The same
results were reported for ΔE obtained from fruit trays as shown in Figures C-4 to C-6. This was
the same pattern for mangos in Figures C-7 to C-12.
The number of reference color or the number of selections of colors did not provide any
information in regard to a correlation between ΔE and CCI. The color change index or measure
of non-uniformity did not have a relationship with the panelist’s performance and error made
when visually evaluating mangos and nectarines. It is possible that above a certain degree of
non-uniformity of colors panelists performance is reduced and becomes too complicated to refer
to color bars.
62
CHAPTER 5 CONCLUSIONS
Colors reflect important quality parameters such as maturity, defects and other color-
dependent attributes. Most agricultural materials e.g. fruits, vegetables, grains, meat, and
seafood have non-uniform shapes, surfaces and colors. It is important to quantify the color
attributes of these materials in order to measure their quality and to help with the record keeping
associated with the new globalization requirements.
The number of color blocks did not provide a clear measure of the degree of non-
uniformity in mangoes and nectarines. However, the number of color primitives associated with
color change index and number of neighbors does provide a better measure of their degree of
non-uniformity.
Quantitative color data can be correlated with human perception. The method developed
in this study can be used to quantify the perceptions of untrained panelists regarding non-
uniformly colored foods, with objective error measurements to optimize the method parameters.
It was observed that there may be an “optimum” number of reference colors for a given food. In
our study, 12 reference colors performed poorly compared to either 8 or 16 reference colors.
Since all of the 8 reference colors were present in the set of 12 reference colors, one may argue
that increasing the number from 8 to 12 “diluted” their effect. This needs to be tested in future
studies. It is more difficult to explain why 16 reference colors performed better than 12
references. Future studies may explore this dilemma.
It was clear that the more color selections, the less the error made by the panelists, given
the reference colors provided in this study. There was also statistically significant interaction
between the number of reference colors and the number of selections. It is possible that up to
certain level of reference colors panelists would perform more efficiently, and above that level it
63
would be too complicated for the panelists to refer to color selections and reference colors. This
issue needs to be elucidated in future studies.
In this study there was no correlation between the error performance of panelists and the
degree of non-uniformity provided by the number of primitives. The concept of the minimum
possible performance level was introduced, the best possible ΔE. This provided a more realistic
way to calculate the error made by sensory panels, given a number of reference colors and color
selections.
Panelists also evaluated the color of the same sample either by looking at its image, or at a
real fruit. This study found small but statistically significant differences in the error made by
panelists between these sources. It is interesting, but expected that the error made when looking
at the image was less, since the reference colors were developed from the images. Specific
studies in the future need to clarify if images can be substituted for the real food, for visual
evaluation purposes.
It is essential to keep identifying criteria to measure the visual evaluations of panelists and
their correlations with instrumental methods of color measurements, to provide a better
understanding to the human perception of non-uniform colors. The search for better methods to
quantify and correlate instrumental and human perception data in this area should continue.
64
APPENDIX A COLOR ANALYSIS FOR ALL TRAYS
Figure A-1. Fruit Tray booth 1 for image acquisition and sensory panel.
Figure A-2. Fruit Tray booth 2 for image acquisition and sensory panel.
65
Figure A-3. Fruit Tray booth 3 for image acquisition and sensory panel.
Figure A-4. Fruit Tray booth 4 for image acquisition and sensory panel.
66
Figure A-5. Fruit Tray booth for image acquisition and sensory panel.
Figure A-6. Fruit Tray booth 6 for image acquisition and sensory panel.
67
Figure A-7. Fruit Tray booth 7 for image acquisition and sensory panel.
Figure A-8. Fruit Tray booth 8 for image acquisition and sensory panel.
68
Figure A-9. Fruit Tray booth 9 for image acquisition and sensory panel.
Figure A-10. Fruit Tray booth 10 for image acquisition and sensory panel.
69
Figure A-11. Machine Vision set-up.
Figure A-12. Light box specifications.
88 cm
46 cm 51 cm
50 cm
70
08
11
01 03 04 05 06 07
12
02 09 10
0601 03 04 05 0802 07
01
11 14
02 04 05 06 07 08 09 10
15 16
03
12 13
8 References
12 References
16 References
Figure A-13. Reference scales presented to panelists.
The main reference color bar was that with 16 colors. From that color bar, color blocks
with close/similar L*a*b or RGB values were merged to reduce the number of colors to choose
and create a color bar with 12 reference colors. The same procedure was used to create the color
reference scale with 8 references colors.
Table A-1. L*a*b values for reference color bar with 8 color. Reference color L* a* b* 1 59.14 -5.61 58.53 2 32.97 56.6 43.38 3 39.89 53.77 34.18 4 52.32 21.89 47.08 5 60.95 2.83 55.43 6 41.77 62.53 46.01 7 71.07 17.04 60.64 8 75.28 7.85 64.73
71
Table A-2. L*a*b values for reference color bar with 12 colors. Reference color L* a* b* 1 59.14 -5.61 58.53 2 32.97 56.6 43.38 3 39.89 53.77 34.18 4 42.15 47.55 36.62 5 52.32 21.89 47.08 6 60.95 2.83 55.43 7 45.33 53.39 40.92 8 41.77 62.53 46.01 9 67.41 28.11 40.77 10 71.07 17.04 60.64 11 79.56 -1.8 74.03 12 75.28 7.85 64.73
Table. A-3 L*a*b values for reference color bar with 16 colors. Reference color L* a* b* 1 50.01 13.55 50.72 2 59.14 -5.61 58.53 3 32.97 56.6 43.38 4 39.89 53.77 34.18 5 42.15 47.55 36.62 6 52.32 21.89 47.08 7 56.51 12.35 51.18 8 60.95 2.83 55.43 9 45.33 53.39 40.92 10 47.94 46.34 43.57 11 41.77 62.53 46.01 12 67.41 28.11 40.77 13 71.07 17.04 60.64 14 79.56 -1.8 74.03 15 75.28 7.85 64.73 16 90.35 -9.87 67.25
72
Figure A-14. Example ballot for screen image evaluation. (Fruit Image Evaluation Form)
Sensory color evaluation form
Date Panelist Age: Male □ Female □ Booth number Instructions: 1. Do not re-orient the samples, or modify their wrapping. 2. Evaluate the samples using the order given here. 3. From the screen, select only 2 colors that best represent the colors of the sample. 4. Estimate the percentage of these colors for the surface of the sample shown. 5. The sum of the 2 percentages must add to 100% Sample number (541) Color number (1 to 8)
Percent of total area
Sum=100% Sample number (397) Color number (1 to 8)
Percent of total area
Sum=100%
73
Figure A-15. Example ballot for fruit evaluation.
(Fruit Tray Evaluation Form)
Sensory color evaluation form Date Panelist Age: Male □ Female □ Booth number Instructions: 1. Do not re-orient the samples, or modify their wrapping. 2. Evaluate the samples using the order given here. 3. From the screen, select only 2 colors that best represent the colors of the sample. 4. Estimate the percentage of these colors for the surface of the sample shown. 5. The sum of the 2 percentages must add to 100% Sample number (397) Color number (1 to 8)
Percent of total area
Sum=100% Sample number (541) Color number (1 to 8)
Percent of total area
Sum=100% Table A-4. Mango color primitives.
Booth CCI # Primitives # Neighbors 1 6.42 616 1190 2 9.21 613 1239 3 5.58 513 982 4 4.90 548 1012 5 6.96 593 1127 6 7.16 615 1190 7 6.76 609 1196 8 4.33 538 964 9 4.58 536 1016 10 11.29 738 1511
74
Table A-5. Nectarine color primitives. Booth CCI # Primitives # Neighbors
1 17.51 915 1760 2 13.72 693 1366 3 13.13 581 1196 4 13.20 794 1521 5 11.32 626 1215 6 14.23 729 1491 7 17.75 856 1648 8 15.19 756 1488 9 11.49 739 1414 10 14.76 773 1566
Figure A-16. Representation of color primitives and equivalent circles for mangos (left) and nectarines (right) with a MV system.
75
APPENDIX B PANELISTS PERFORMANCE FOR MANGOS AND NECTARINES
10. Treatment 1 = 8 reference color and 2 color selections
11. Treatment 2 = 8 reference colors and 4 color selections
12. Treatment 3 = 8 reference colors and 6 color selections
13. Treatment = 12 reference color and 2 color selections
14. Treatment 5 = 12 reference colors and 4 color selections
15. Treatment 6 = 12 reference colors and 6 color selections
16. Treatment 7 = 16 reference colors and 2 color selections
17. Treatment 8 = 16 reference colors and 4 color selections
18. Treatment 9 = 16 reference colors and 6 color selections
Table B-1. Summary performance for panelists evaluating both fruits for booth 1. Nectarine Mango
Case Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
1* 1.34 9.15 4.79 12.97 8.38 2.92 11.39 9.56 15.75 6.05 2* 2.82 7.41 2.79 11.31 3.52 2.79 10.70 5.16 10.62 2.30 3* 0.20 7.61 2.72 10.09 2.38 0.20 9.49 4.04 10.52 5.41 4* 1.34 25.73 14.36 20.02 14.59 2.92 21.15 7.64 18.04 8.86 5* 0.20 13.54 9.16 11.07 1.26 0.20 13.53 7.79 12.84 3.01 6* 0.20 8.77 1.82 10.61 4.04 0.20 13.87 8.26 12.73 4.74 7* 1.34 13.27 11.20 12.66 1.42 2.92 15.20 10.29 14.93 4.89 8* 0.20 9.95 6.33 12.57 3.76 0.20 13.97 9.51 11.34 1.86 9* 0.20 14.13 10.59 10.57 4.92 0.20 13.91 5.91 6.40 6.40
76
Table B-2. Summary performance for panelists evaluating both fruits for booth 2. Nectarine Mango
Case Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
1* 0.90 11.91 5.20 9.46 3.55 1.90 11.04 7.00 15.36 5.87 2* 2.56 6.36 3.27 10.02 1.88 1.38 9.15 5.23 10.02 4.12 3* 0.20 10.44 9.88 6.10 1.35 0.20 10.51 5.77 11.13 3.95 4* 0.91 14.93 7.46 21.29 16.93 1.90 15.75 6.93 20.84 7.87 5* 0.20 8.73 2.85 11.01 2.34 0.20 9.53 3.63 10.97 4.69 6* 0.19 6.09 3.49 10.23 4.23 1.90 8.78 5.48 11.73 5.00 7* 1.34 13.36 4.59 13.09 4.79 2.92 15.50 8.52 15.15 7.15 8* 0.20 9.19 1.31 11.63 4.17 0.20 8.60 8.60 13.50 6.62 9* 0.20 12.49 7.31 10.77 2.24 0.20 8.60 3.01 12.10 12.10
Table B-3. Summary performance for panelists evaluating both fruits for booth 3.
Nectarine Mango
Case Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
1* 1.90 9.15 4.72 7.69 3.14 2.40 10.72 6.26 10.85 1.46 2* 2.13 7.94 4.46 8.67 3.54 0.18 9.15 5.23 10.02 4.12 3* 0.20 5.39 2.81 5.73 2.10 0.20 10.59 5.09 8.34 4.48 4* 1.89 20.09 14.33 16.52 12.38 2.41 16.55 11.69 9.90 3.52 5* 0.20 14.67 8.71 10.68 8.71 0.20 12.31 7.82 12.16 4.68 6* 0.20 7.60 2.82 7.35 4.61 0.20 7.87 4.08 7.54 3.42 7* 1.90 9.79 5.82 8.58 3.40 2.41 11.48 6.38 15.47 7.19 8* 0.20 12.74 8.25 12.43 8.25 0.20 7.82 5.86 10.28 2.63 9* 0.20 7.79 3.97 7.38 3.66 0.20 9.46 8.39 9.69 9.69
Table B-4. Summary performance for panelists evaluating both fruits for booth 4.
Nectarine Mango
Case Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
1* 4.66 11.91 7.34 14.87 4.70 2.59 7.19 4.71 15.59 5.49 2* 5.93 14.46 6.65 13.98 4.70 2.58 11.44 3.71 13.31 3.15 3* 0.20 14.16 9.21 12.78 2.07 0.20 12.05 7.14 15.14 3.94 4* 4.66 28.17 11.16 14.58 10.72 2.59 19.67 8.85 23.81 9.87 5* 0.20 14.08 7.71 14.04 3.64 2.35 10.76 5.43 18.19 6.04 6* 0.20 10.45 5.21 13.18 1.82 0.20 11.58 5.65 11.03 6.09 7* 4.66 14.66 7.97 17.08 7.40 2.59 11.06 5.31 18.05 9.57 8* 0.20 13.44 6.36 16.89 6.36 0.20 12.74 7.01 18.41 2.38 9* 0.20 9.14 3.68 17.39 7.27 0.20 11.33 4.93 15.09 15.09
77
Table B-5. Summary performance for panelists evaluating both fruits for booth 5. Nectarine Mango
Case Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
1* 2.65 10.23 5.71 9.92 2.64 1.93 11.98 6.53 19.10 5.68 2* 8.81 11.77 10.50 11.75 7.95 1.79 12.94 5.21 11.84 5.07 3* 0.20 6.44 2.54 11.61 6.57 0.20 9.26 4.48 12.30 4.49 4* 2.65 15.11 3.07 13.43 5.00 1.93 22.01 9.40 21.45 7.65 5* 0.20 11.93 5.84 17.36 8.56 1.72 12.30 6.52 15.76 5.79 6* 0.20 6.76 2.52 13.84 6.81 0.20 10.32 4.37 15.52 4.24 7* 2.65 15.25 9.75 18.36 7.43 1.93 12.07 2.42 17.39 7.39 8* 0.20 14.04 8.55 13.15 8.55 0.20 15.73 10.68 14.64 3.05 9* 0.20 9.71 4.51 14.16 5.20 0.20 13.78 4.57 16.04 16.04
Table B-6. Summary performance for panelists evaluating both fruits for booth 6. Nectarine Mango
Case Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
1* 1.27 13.24 6.92 16.02 9.33 2.67 13.15 7.69 19.44 10.38 2* 6.85 11.48 6.36 12.39 4.91 2.99 11.07 5.56 15.20 8.04 3* 0.20 13.13 6.10 11.55 3.20 0.20 8.48 2.84 10.01 3.32 4* 1.27 20.59 5.85 25.95 12.12 2.67 16.86 4.54 21.79 7.02 5* 0.20 7.17 2.76 8.59 2.03 0.20 9.50 3.97 11.22 4.53 6* 0.20 7.28 1.76 7.30 4.76 0.20 11.58 5.65 11.03 6.09 7* 1.27 12.37 8.48 10.63 2.21 2.67 12.07 2.42 17.39 7.39 8* 0.20 9.02 3.37 8.57 3.37 0.20 12.79 5.06 12.15 3.88 9* 0.20 8.29 2.40 9.63 3.65 0.20 10.84 6.06 10.17 10.17
Table B-7. Summary performance for panelists evaluating both fruits for booth 7. Nectarine Mango
Case Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
1* 8.67 13.75 8.90 17.46 9.89 2.19 12.66 9.72 11.80 2.94 2* 11.59 15.28 5.80 15.94 9.54 0.66 13.27 6.68 9.00 2.91 3* 0.20 12.70 3.43 13.13 2.18 0.20 5.19 1.97 8.21 0.92 4* 8.67 23.78 7.06 22.47 8.77 2.19 17.21 17.44 15.84 16.89 5* 0.20 16.34 8.74 18.53 5.54 0.20 13.03 11.79 12.28 10.80 6* 0.20 14.43 7.37 18.62 7.86 0.20 11.98 9.70 11.88 6.16 7* 8.67 13.29 5.03 16.91 7.01 2.19 11.54 5.54 11.78 2.80 8* 0.20 18.12 8.37 16.39 8.37 0.20 8.02 4.94 11.22 5.19 9* 0.20 11.93 5.78 16.85 4.13 0.20 10.48 12.83 11.30 11.30
78
Table B-8. Summary performance for panelists evaluating both fruits for booth 8. Nectarine Mango
Case Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
1* 1.25 11.28 9.21 15.52 5.38 0.83 13.25 11.45 13.42 4.07 2* 3.69 7.14 4.47 9.59 2.15 0.69 9.47 4.16 12.05 2.69 3* 0.20 7.68 3.08 9.73 3.76 0.20 10.39 3.14 10.90 3.80 4* 1.25 21.03 13.05 24.14 16.85 0.83 23.58 15.44 11.13 12.015* 0.20 9.49 5.06 11.96 2.62 0.20 14.69 10.29 12.15 3.50 6* 0.20 11.31 7.19 8.95 3.14 0.20 12.44 7.46 10.87 3.56 7* 1.25 16.38 6.68 20.19 7.65 0.83 9.87 4.58 13.43 7.18 8* 0.20 4.86 2.30 13.65 2.30 0.20 11.92 5.69 13.10 3.36 9* 0.20 8.69 3.74 14.58 4.33 0.20 11.44 8.81 13.76 13.76
Table B-9. Summary performance for panelists evaluating both fruits for booth 9. Nectarine Mango
Case Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
1* 4.65 8.95 6.12 12.86 3.66 0.61 8.77 7.13 12.83 3.58 2* 5.45 10.38 4.01 14.28 7.35 0.59 10.69 4.37 13.70 4.18 3* 0.20 8.17 2.08 11.40 3.30 0.20 11.47 3.66 10.53 2.95 4* 4.65 24.19 10.96 19.32 13.82 0.61 18.88 11.63 35.88 15.715* 0.20 13.35 4.51 15.40 4.91 0.20 13.01 4.73 13.59 4.26 6* 0.20 12.02 3.60 11.06 2.69 0.20 13.01 4.73 13.59 4.26 7* 4.65 13.07 5.34 21.78 8.71 0.61 10.87 7.18 11.36 4.55 8* 0.20 12.68 6.82 13.33 6.82 0.20 11.08 7.83 11.88 4.45 9* 0.20 11.07 3.57 14.15 4.07 0.20 10.49 5.53 16.38 16.38
Table B-10. Summary performance for panelists evaluating both fruits for booth 10. Nectarine Mango
Case Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
Best ΔE
Screen ΔE Stdev
Tray ΔE Stdev
1* 7.08 9.23 6.05 17.59 4.57 2.80 7.27 4.89 11.59 3.74 2* 7.22 13.13 5.15 15.19 4.93 2.18 10.13 9.04 12.65 5.37 3* 0.20 12.72 3.53 17.77 5.69 0.20 9.00 4.78 12.68 6.86 4* 7.10 27.25 10.32 26.93 10.90 2.80 14.39 8.22 17.80 13.21 5* 0.20 12.79 3.99 15.93 2.94 0.20 15.16 6.04 11.19 3.60 6* 0.20 9.58 2.65 15.96 5.15 0.20 9.93 4.79 10.14 3.01 7* 7.08 18.04 5.65 21.21 7.77 2.80 12.83 7.72 16.43 5.28 8* 0.20 10.86 4.36 14.97 4.36 0.20 5.99 1.54 10.66 2.71 9* 0.20 9.92 5.22 15.42 3.01 0.20 7.18 2.59 10.91 10.91
79
APPENDIX C ΔE VS CCI FOR ALL COMBINATIONS
11 12 13 14 15 16 17 18 19CCI
0
10
20
30
40D
elta
E
8 ref, 2 8 ref, 4 8 ref, 6
Nectarine, abs.DE, screen
Figure C-1. Absolute Δ E for nectarine for screen image and 8 references.
11 12 13 14 15 16 17 18 19CCI
0
10
20
30
40
Del
ta E
12 ref, 2 12 ref, 4 12 ref 6
Nectarine, abs.DE, screen
Figure C-2. Absolute Δ E for nectarine for screen image and 12 references.
80
11 12 13 14 15 16 17 18 19CCI
0
10
20
30
40
Del
ta E
16 ref, 2 16 ref, 4 16 ref, 6
Nectarine, abs.DE, screen
Figure C-3. Absolute Δ E for nectarine for screen image and 16 references.
11 12 13 14 15 16 17 18 19CCI
0
10
20
30
40
Del
ta E
8ref, 2 8 ref, 4 8 ref, 6
Nectarine, abs.DE, fruit
Figure C-4. Absolute Δ E for nectarine for tray and 8 references.
81
11 12 13 14 15 16 17 18 19CCI
0
10
20
30
40
Del
ta E
12 ef, 2 12 ref, 4 12 ref, 6
Nectarine, abs.DE, fruit
Figure C-5. Absolute Δ E for nectarine for tray and 12 references.
11 12 13 14 15 16 17 18 19CCI
0
10
20
30
40
Del
ta E
16 ref, 2 16 ref, 4 16 ref, 6
Nectarine, abs.DE, fruit
Figure C-6. Absolute Δ E for nectarine for tray and 16 references.
82
3 4 5 6 7 8 9 10 11 12CCI
0
10
20
30
40D
elta
E8ref, 2 8 ref, 4 8 ref, 6
Mango, abs.DE, screen
Figure C-7. Absolute Δ E for mango for screen image and 8 references.
3 4 5 6 7 8 9 10 11 12CCI
0
10
20
30
40
Del
ta E
12 ef, 2 12 ref, 4 12 ref, 6
Mango, abs.DE, screen
Figure C-8. Absolute Δ E for mango for screen image and 12 references.
83
3 4 5 6 7 8 9 10 11 12CCI
0
10
20
30
40D
elta
E16 ref, 2 16 ref, 4 16 ref, 6
Mango, abs.DE, screen
Figure C-9. Absolute Δ E for mango for screen image and 16 references.
3 4 5 6 7 8 9 10 11 12CCI
0
10
20
30
40
Del
ta E
8ref, 2 8 ref, 4 8 ref, 6
Mango, abs.DE, fruit
Figure C-10. Absolute Δ E for mango for tray 8 references.
84
3 4 5 6 7 8 9 10 11 12CCI
0
10
20
30
40
Del
ta E
12 ef, 2 12 ref, 4 12 ref, 6
Mango, abs.DE, fruit
Figure C-11. Absolute Δ E for mango for tray and 12 references.
3 4 5 6 7 8 9 10 11 12CCI
0
10
20
30
40
Del
ta E
16 ref, 2 16 ref, 4 16 ref, 6
Mango, abs.DE, fruit
Figure C-12. Absolute Δ E for mango for tray and 16 references.
85
APPENDIX D DELTA E VALUES FOR DIFFERENT CASES
0 1 2 3 4 5 6 7 8 9 10 11trays
0
10
20
30
40D
elta
E8 ref, 28 ref, 48 ref, 6
Nectrarine, abs.DE, screen
Figure D-1. Absolute Δ E for nectarine for screen image and 8 references.
0 1 2 3 4 5 6 7 8 9 10 11Trays
0
10
20
30
40
Del
ta E
12 ref, 2 12 ref, 4 12 ref 6
Nectarine, Abs.DE, screen
Figure D-2. Absolute Δ E for nectarine for screen image and 12 references.
86
0 1 2 3 4 5 6 7 8 9 10 11Trays
0
10
20
30
40
Del
ta E
16 ref, 2 16 ref, 4 16 ref, 6
Nectarine, abs.DE, screen
Figure D-3. Absolute Δ E for nectarine for screen image and 16 references.
0 1 2 3 4 5 6 7 8 9 10 11trays
0
10
20
30
40
Del
ta E
8 ref, 2 8 ref, 4 8 ref, 6
Nectarine, abs.DE, real fruit
Figure D-4. Absolute Δ E for nectarine for tray and 8 references.
87
0 1 2 3 4 5 6 7 8 9 10 11trays
0
10
20
30
40
Del
ta E
12 ref, 2 12 ref, 4 12 ref, 6
Nectarine, abs.DE, real fruit
Figure D-5. Absolute Δ E for nectarine for tray and 12 references.
0 1 2 3 4 5 6 7 8 9 10 11trays
0
10
20
30
40
Del
ta E
16 ref, 2 16 ref, 4 16 ref, 6
Nectarine, abs.DE, real fruit
Figure D-6. Absolute Δ E for nectarine for tray and 16 references.
88
0 1 2 3 4 5 6 7 8 9 10 11trays
0
10
20
30
40
Del
ta E
8 ref, 2 8 ref, 4 8 ref, 6
Mango, abs.DE, screen
Figure D-7. Absolute Δ E for mango for screen image and 8 references.
0 1 2 3 4 5 6 7 8 9 10 11trays
0
10
20
30
40
Del
ta E
12 ref, 2 12 ref, 4 12 ref, 6
Mango, abs.DE, screen
Figure D-8. Absolute Δ E for mango for screen image and 12 references.
89
0 1 2 3 4 5 6 7 8 9 10 11trays
0
10
20
30
40
Del
ta E
16 ref, 2 16 ref, 4 16 ref, 6
Mango, abs.DE, screen
Figure D-9. Absolute Δ E for mango for screen image and 16 references.
0 1 2 3 4 5 6 7 8 9 10 11trays
0
10
20
30
40
Del
ta E
8 ref, 2 8 ref, 4 8 ref, 6
Mango, abs.DE, real fruit
Figure D-10. Absolute Δ E for mango for tray 8 references.
90
0 1 2 3 4 5 6 7 8 9 10 11trays
0
10
20
30
40
Del
ta E
12 ref, 2 12 ref, 4 12 ref, 6
Mango, abs.DE, real fruit
Figure D-11. Absolute Δ E for mango for tray and 12 references.
0 1 2 3 4 5 6 7 8 9 10 11trays
0
10
20
30
40
Del
ta E
16 ref, 2 16 ref, 4 16 ref, 6
Mango, abs.DE, real fruit
Figure D-12. Absolute Δ E for mango for tray and 16 references.
91
APPENDIX E SOURCE CODES FOR SAS PROGRAMS
Method 1.
PROC PRINT DATA=FILE; RUN; proc sort data= File; by Object; proc anova data=file; by Object; class reference_colors selections source panelist booth; model Delta_E = reference_colors|selections|source; means reference_colors selections source reference_colors|selections|source/duncan; run;
Method 2.
proc sort data= File; by object; proc mixed DATA=File; by object; class reference_colors selections source booth panelist; model DiffDE = reference_colors|selections|source; random panelist booth; lsmeans selections|reference_colors|source /pdiff; run;
*** The model statement was interchangeable to Diff ΔE or ΔE to statistically analyze both dependent variables. Table E-1. Mixed mode summary absolute ΔE for mangos.
Source Num. DF*
Den. DF*
F Value Pr >F
Reference colors 2 1343 23.21 < .0001 Selections 2 1343 48.73 < .0001 Presentation 1 1343 23.32 < .0001 Reference colors * selections 4 1343 10.76 < .0001 Reference colors* presentation 2 1343 2.09 0.124 Selections*presentation 2 1343 1.51 0.222 Reference colors*selections*presentation
4 1343 0.36 0.835
* Num DF refers to numerator degrees of freedom. Den DF refers to denominator degrees of freedom.
92
Table E-2. Mixed mode summary difference ΔE for mangos. Source Num.
DF* Den. DF*
F Value Pr >F
Reference colors 2 1413 28.57 < .0001 Selections 2 1413 14.90 < .0001 Presentation 1 1413 23.17 < .0001 Reference colors * selections 4 1413 9.71 < .0001 Reference colors* presentation 2 1413 2.08 0.126 Selections*presentation 2 1413 1.50 0.224 Reference colors*selections*presentation
4 1413 0.36 0.837
* Num DF refers to numerator degrees of freedom. Den DF refers to denominator degrees of freedom. Table E-3. Mixed Mode summary absolute ΔE for nectarines.
Source Num. DF*
Den. DF*
F Value Pr >F
Reference colors 2 1343 32.84 < .0001 Selections 2 1343 88.95 < .0001 Presentation 1 1343 17.70 < .0001 Reference colors * selections 4 1343 22.37 < .0001 Reference colors* presentation 2 1343 1.79 0.167 Selections*presentation 2 1343 1.43 0.241 Reference colors*selections*presentation
4 1343 1.38 0.237
* Num DF refers to numerator degrees of freedom. Den DF refers to denominator degrees of freedom. Table E-4. Mixed Mode summary difference in ΔE for nectarines.
Source Num. DF*
Den. DF*
F Value Pr >F
Reference colors 2 1413 77.58 < .0001 Selections 2 1413 21.20 < .0001 Presentation 1 1413 16.96 < .0001 Reference colors * selections 4 1413 25.41 < .0001 Reference colors* presentation 2 1413 1.72 0.126 Selections*presentation 2 1413 1.37 0.224 Reference colors*selections*presentation
4 1413 1.33 0.837
* Num DF refers to numerator degrees of freedom. Den DF refers to denominator degrees of freedom.
93
The Mixed Procedure Type 3 Tests of Fixed Effects Num Den Effect DF DF F Value Pr > F Selections*Source 2 1343 1.51 0.2219 Refere*Select*Source 4 1343 0.36 0.8353 Least Squares Means Reference Standard Effect Source Colors Selections Estimate Error DF t Value Selections 2 15.2401 0.5072 1343 30.05 Selections 4 11.9144 0.5072 1343 23.49 Selections 6 11.2349 0.5072 1343 22.15 Least Squares Means Reference Effect Source Colors Selections Pr > |t| Selections 2 <.0001 Selections 4 <.0001 Selections 6 <.0001 Differences of Least Squares Means Standard Effect Selections Selections Estimate Error DF t Value Pr > |t| Selections 2 4 3.3258 0.4342 1343 7.66 <.0001 Selections 2 6 4.0053 0.4342 1343 9.22 <.0001 Selections 4 6 0.6795 0.4342 1343 1.56 0.1179
Figure E-1. Absolute ΔE means for selection of color for mangos.
The Mixed Procedure Type 3 Tests of Fixed Effects Num Den Effect DF DF F Value Pr > F Refere*Select*Source 4 1413 0.36 0.8369 Least Squares Means Standard Effect Selections Estimate Error DF t Value Pr > |t| Selections 2 13.1554 0.4968 1413 26.48 <.0001 Selections 4 11.1637 0.4968 1413 22.47 <.0001 Selections 6 11.0349 0.4968 1413 22.21 <.0001 Differences of Least Squares Means Standard Effect Selections Selections Estimate Error DF t Value Pr > |t| Selections 2 4 1.9917 0.4356 1413 4.57 <.0001 Selections 2 6 2.1205 0.4356 1413 4.87 <.0001 Selections 4 6 0.1288 0.4356 1413 0.30 0.7675
Figure E-2. Difference ΔE Means for selection of color for mangos.
94
The Mixed Procedure Type 3 Tests of Fixed Effects Least Squares Means Standard Effect Selections Estimate Error DF t Value Pr > |t| Reference_Colors 8 11.4803 0.5072 1343 22.64 <.0001 Reference_Colors 12 14.3974 0.5072 1343 28.39 <.0001 Reference_Colors 16 12.5116 0.5072 1343 24.67 <.0001
Differences of Least Squares Means
Standard Effect Selections Estimate Error DF t Value Pr > |t| Reference_Colors 8 12 ‐2.9171 0.4342 1343 ‐6.72 <.0001 Reference_Colors 8 16 ‐1.0314 0.4342 1343 ‐2.38 0.0177 Reference_Colors 12 16 1.8858 0.4342 1343 4.34 <.0001
Figure E-3. Absolute ΔE means for reference colors for mangos.
The Mixed Procedure Type 3 Tests of Fixed Effects Least Squares Means Standard Effect Selections Estimate Error DF t Value Pr > |t| Reference_Colors 8 10.1910 0.4968 1343 20.52 <.0001 Reference_Colors 12 13.4795 0.4968 1343 27.14 <.0001 Reference_Colors 16 11.6834 0.4968 1343 23.52 <.0001
Differences of Least Squares Means Standard Effect Selections Estimate Error DF t Value Pr > |t| Reference_Colors 8 12 ‐3.2885 0.4356 1343 ‐7.55 <.0001 Reference_Colors 8 16 ‐1.4924 0.4356 1343 ‐3.43 0.0006 Reference_Colors 12 16 1.7961 0.4356 1343 4.12 <.0001
Figure E-4. Difference ΔE means for reference colors for mangos.
The Mixed Procedure Type 3 Tests of Fixed Effects Least Squares Means Standard Effect Selections Estimate Error DF t Value Pr > |t| Source F 13.6525 0.4752 1343 28.73 <.0001 Source S 11.9404 0.4752 1343 25.13 <.0001
Differences of Least Squares Means Standard Effect Selections Estimate Error DF t Value Pr > |t| Source F S 1.7122 0.3546 1343 4.83 <.0001
Figure E-5. Absolute ΔE means for presentation for mangos.
95
The Mixed Procedure Type 3 Tests of Fixed Effects Least Squares Means Standard Effect Selections Estimate Error DF t Value Pr > |t| Source F 12.6407 0.4638 1343 27.25 <.0001 Source S 10.9286 0.4638 1343 23.56 <.0001
Differences of Least Squares Means
Standard Effect Selections Estimate Error DF t Value Pr > |t| Source F S 1.7121 0.3557 1343 4.81 <.0001
Figure E-6. Difference ΔE means for presentation for mangos.
The Mixed Procedure Type 3 Tests of Fixed Effects Least Squares Means Reference Standard Effect Source Colors Selections Estimate Error DF t Value Selections 2 16.3154 0.7332 1343 22.25 Selections 4 12.2299 0.7332 1343 16.68 Selections 6 10.8946 0.7332 1343 14.86 Least Squares Means Reference Effect Source Colors Selections Pr > |t| Selections 2 <.0001 Selections 4 <.0001 Selections 6 <.0001 Differences of Least Squares Means Standard Effect Selections Selections Estimate Error DF t Value Pr > |t| Selections 2 4 4.0855 0.4235 1413 9.65 <.0001 Selections 2 6 5.4208 0.4235 1413 12.80 <.0001 Selections 4 6 1.3353 0.4235 1413 3.15 0.0016 Figure E-7. Absolute ΔE means for selections of colors for nectarines.
96
The Mixed Procedure Type 3 Tests of Fixed Effects Least Squares Means Standard Effect Selections Estimate Error DF t Value Pr > |t| Selections 2 12.8803 0.4604 1413 27.97 <.0001 Selections 4 10.2490 0.4604 1413 22.26 <.0001 Selections 6 10.6946 0.4604 1413 23.23 <.0001 Differences of Least Squares Means Standard Effect Selections Selections Estimate Error DF t Value Pr > |t| Selections 2 4 2.6313 0.4326 1413 6.08 <.0001 Selections 2 6 2.1857 0.4326 1413 5.05 <.0001 Selections 4 6 ‐0.4456 0.4326 1413 ‐1.03 0.3031 Figure E-8. Difference ΔE means for selections of colors for nectarines.
The Mixed Procedure Type 3 Tests of Fixed Effects Least Squares Means Standard Effect Selections Estimate Error DF t Value Pr > |t| Reference_Colors 8 11.4792 0.7332 1343 15.66 <.0001 Reference_Colors 12 14.9077 0.7332 1343 20.33 <.0001 Reference_Colors 16 13.0530 0.7332 1343 17.80 <.0001
Differences of Least Squares Means Standard Effect Selections Estimate Error DF t Value Pr > |t| Reference_Colors 8 12 ‐3.4285 0.4235 1343 ‐8.10 <.0001 Reference_Colors 8 16 ‐1.5737 0.4345 1343 ‐3.72 0.0002 Reference_Colors 12 16 1.8547 0.4345 1343 4.38 <.0001
Figure E-9. Absolute ΔE means for reference colors for nectarines.
The Mixed Procedure Type 3 Tests of Fixed Effects Least Squares Means Standard Effect Selections Estimate Error DF t Value Pr > |t| Reference_Colors 8 8.3653 0.4604 1343 18.17 <.0001 Reference_Colors 12 13.6839 0.4604 1343 29.72 <.0001 Reference_Colors 16 11.7746 0.4604 1343 25.57 <.0001
Differences of Least Squares Means Standard Effect Selections Estimate Error DF t Value Pr > |t| Reference_Colors 8 12 ‐5.3186 0.4326 1343 ‐12.29 <.0001 Reference_Colors 8 16 ‐3.4092 0.4326 1343 ‐7.88 <.0001 Reference_Colors 12 16 1.9093 0.4326 1343 4.41 <.0001
Figure E-10. Difference ΔE means for reference colors for nectarines.
97
The Mixed Procedure Type 3 Tests of Fixed Effects Least Squares Means Standard Effect Selections Estimate Error DF t Value Pr > |t| Source F 13.8739 0.7125 1343 19.47 <.0001 Source S 12.4193 0.7125 1343 17.43 <.0001
Differences of Least Squares Means
Standard Effect Selections Estimate Error DF t Value Pr > |t| Source F S 1.4546 0.3458 1343 4.21 <.0001
Figure E-11. Absolute ΔE means for presentation for nectarines.
The Mixed Procedure Type 3 Tests of Fixed Effects Least Squares Means Standard Effect Selections Estimate Error DF t Value Pr > |t| Source F 12.0020 0.4252 1343 28.23 <.0001 Source S 10.5473 0.4252 1343 24.81 <.0001
Differences of Least Squares Means
Standard Effect Selections Estimate Error DF t Value Pr > |t| Source F S 1.4547 0.3532 1343 4.12 <.0001
Figure E-12. Difference ΔE means for presentation for nectarines.
98
LIST OF REFERENCES
Abdullah ZM, Aziz AS, Dos-Mohamed AM. 2000. Quality Inspection of Bakery Products Using a Color Based Machine Vision System. Journal of Food Quality 23(1):39-50.
Balaban MO. 2007. Quantifying Non-Homogeneous Colors in Agricultural Materials. Part 1:
Method Development. Journal of Food Science. Submitted July 2007. p 1-21. Balaban MO, Aparicio J, Zotarelli M, Sims C. 2007. Quantifying non-homogeneous colors in
agricultural materials. Part II: comparison of machine vision and sensory panel evaluations. Journal of Food Science. Submitted July 2007. p 1-20.
Balaban MO, Odabasi AZ. 2006 Color Measurement with Machine Vision. Food
Technology:32-6. Basset O, Buquet B, Abouelkaram S, Delachartre P, Culioli J. 2000. Application of Texture
Image Analysis for the Classification of Bovine Meat. Journal of Food Chemistry 69(4):437-45.
Bertrand D, Le Guerneve C, Marion D, Devaux MF, Robert P. 1992. Description of the Textural
Appearance of Bread Crumb Appearance by Video Image Analysis. Cereal Chemistry 69(3):257-61.
Bharati MH, Jay Liu J, MacGregor JF. 2004. Image Texture Analysis: Methods and
Comparisons. Chemometrics and Intelligence Laboratory Systems 72:57-71. Blasco J, Aleixos N, Molto E. 2003. Machine Vision System for Automatic Quality Grading of
Fruit. Biosystems Engineering 85(4):415-23. Brewer MS, Zhu LG, Bidner B, Meisinger DJ, McKeith FK. 2001. Measuring Pork Color:
Effects of Bloom, Time, Muscle pH and Relationship to Instrumental Parameters. Meat Science 57(2):169-76.
Brosnan T, Sun D-W. 2004. Improving Quality Inspection of Food Products by Computer Vision
- A Review. Journal of Food Engineering 61:3-16. Chalidabhongse T, Yimyam P, Sirisomboon P. 2006. 2D/3D Vision-Based Mango's Feature
Extraction and Sorting. 9th International Conference on Control, Automation, Robotics and Vision. December 5-8; Singapore.
Chandraratne MR, Samarasinghe S, Kulasiri D, Bickerstaffe R. 2006. Prediction of Lamb
Tenderness Using Image Surface Texture Features. Journal of Food Engineering 77:492-9.
Chapman FA, Fitz-Coy SA, Thunberg EM, M. AC. 1997. United States Trade in Ornamental
Fish. Journal of the World Aquaculture Society 28:1-10.
99
Chizzolini R, Badiani A, Rosa P, Novelli E.1993. Objective and Sensorial Evaluation of Pork Quality: A Comprehensive Study. Proceedings for 39th International Congress of Meat Science and Technology. Calgary, Alberta, California. p 185.
Coles G, Lammerink JP, Wallace AR. 1993. Estimating Potato Crisp Colour Variability Using
Image Analysis and a Quick Visual Method. Potato Research 36:127-34. Cornforth D. 1994. Quality Attributes and Their Measurement in Meat, Poultry and Fish
Products. New York: Blackie Academic & Professional. Davidson VJ, Ryks J, Chu T. 2001. Fuzzy Models to Predict Consumer Ratings for Biscuits
Based on Digital Features. IEEE Transactions on Fuzzy Systems 9(1):62-7. Delwiche MJ. 1987. Grader performance using a peach ground color maturity chart. Journal of
the American Society for Horticultural Science 22(1):87-9. Denoyelle C, Berny F. 1999. Objective Measurement of Veal Color for Classification Purposes.
Meat Science 53:203-9. Du CJ, Sun DW. 2004. Recent Developments In The Applications Of Image Processing
Techniques for Food Quality Evaluation. Trends in Food Science & Technology 15:230-49.
FAO. 2006. FAO Newsroom: Nearly Half of all Fish Eaten Today Farmed, Not Caught.
Available from: http://www.fao.org/newsroom/en/news/2006/1000383/index.html. Accessed 08-24-07
Fernandez L, Castillero C, Aguilera JM. 2005. An Application of Image Analysis to Dehydration
of Apple Discs. Journal of Food Engineering 67(1-2):185-93. Gerrard DE, Gao X, Tan J. 1996. Beef Marbling and Color Score Determination by Image
Processing. Journal of Food Science 61(1):145-8. Gunasekaran S. 1996. Computer Vision Technology for Food Quality Assurance. Trends in Food
Science & Technology 7(8):245-56. Hatano M, Takahashi K, Onishi A, Kameyama Y. 1989. Quality Standardization of Fall Chum
Salmon by Digital Image Processor. Nippon Suisan Gakkaishi 55:1427–33. Hatem I, Tan J, Gerrard DE. 2003. Determination of Animal Skeletal Maturity by Image
Processing. Journal of Meat Science 65:999-1004. Hayashi S, Kanuma T, Ganno K, Sakakue O. 1998. Cabbage Head Recognition and Size
Estimation for Development of a Selective Harvester. ASAE Annual International Meeting; Michigan Paper No. 983042.
100
Hienemann PH, Hughes R, Morrow CT, Sommer HJ, Beelman RB, Wuest PJ. 1994. Grading of Mushrooms Using a Machine Vision System. Transactions of the ASAE 37(5):1671-7.
Howarth MS, Searcy SW. 1992. Inspection of Fresh Carrots by Machine Vision. Food
Processing Automation II Proceedings of the 1992 Conference; St. Joseph, Michigan USA: ASAE.
Kane AM, Lyon BG, Swanson RB, Savage EM. 2003. Comparison of Two Sensory and Two
Instrumental Methods to Evaluate Cookie Color. Journal of Food Science 68(5):1831-7. Kartikeyan B, Sarkar A. 1991. An Identification Approach for 2-D Autoregressive Models in
Describing Textures. Graphical Models and Image Processing 53:121-31. Kondo N, Ahmad U, Monta M, Murasc H. 2000. Machine Vision Based Quality Evaluation of
Iyokan Orange Fruit Using Neural Networks. Computers and Electronics in Agriculture 29(1-2):135-47.
Korel F, Luzuriaga DA, Balaban MO. 2001. Quality Evaluation of Raw and Cooked Catfish
(Ictalurus punctatus) Using Eletronic Nose and Machine Vision. Journal of Aquatic Food Product Technology 10(1):3-18.
Lee DJ, Redd S, Schoenberger R, Xu X, Zhan P. 2003. An Automated Fish Species
Classification and Migration Monitoring System. Proceedings for The 29th Annual Conference of The IEEE Industrial Electronics Society November 2-6; Roanoke, Virginia. p 1080-5.
Lee DJ, Xu X, Lane RM, Zhan P. 2004. Shape Analysis for an Automatic Oyster Grading
System. SPIE Optics East, Two and Three-Dimensional Vision Systems for Inspection, Control, and Metrology II October 25-28; Philadelphia, PA, USA.
Leemans V, Magein H, Destain MF. 1998. Defects Segmentation on "Golden Delicious" apples
by Using Colour Machine Vision. Computers and Electronics in Agriculture 20:117-30. Li J, Tan J, Martz FA. 1997. Predicting Beef Tenderness From Image Texture Features. ASAE
Annual International Meeting; St. Joseph, Michigan , USA Paper No. 973124. Liu J, Paulsen MR. 1997. Corn Whiteness Measurement and Classification Using Machine
Vision. 1997 ASAE Annual International Meeting St. Joseph, Michigan, USA:ASAE. Paper No. 973045.
Lou X, Jayas DS, Symons SJ. 1999. Identification of Damaged Kernels in Wheat Using a Color
Machine Vision System. Journal of Cereal Science 30:49-59. Louka N, Juhel F, Fazilleau V, Loonis P. 2004. A Novel Colorimetry Analysis Used to Compare
Different Drying Fish Processes. Food Control 15:327-34.
101
Lu J, Tan J, Shatadal P, Gerrard DE. 2000. Evaluation of Pork Color by Using Computer Vision. Journal of Meat Science 56(1):57-60.
Mäenpää T. 2003. The Local Binary Pattern Approach to Texture Analysis-Extensions and
Applications.PhD. Dissertation. Infotech Oulu and Department of Electrical and Information Engineering, University of Oulu.Oulu, Finland.
Majumdar S, Jayas DS. 2000. Classification of Cereal Grains Using Machine Vision: I
Morphology Models. Transactions of The ASAE 43(6):1669-75. Mancini RA, Hunt MC. 2005. Current Research in Meat Color. Journal of Muscle Foods
71:100-21. Mao J, Jain AK. 1992. Texture Classification and Segmentation Using Multi-resolution
Simultaneous Autoregressive Models. Pattern Recognition 25(2):173-88. Martin A, Tosunoglu S. 2000. Image Processing Techniques For Machine Vision. Florida
Conference on Recent Advances in Robotics; Boca Raton, FL: Florida Atlantic University.
Marty-Mahe P, Loisel P, Fauconneau B, Haffray P, Brossard D, Davenel A. 2004. Quality Traits
of Brown Trouts (Salmo trutta) Cutlets Described by Automated Color Image Analysis. Aquaculture 232:225-40.
Mertens K, De Ketelaere B, Kamers B, Bamelis FR, Kemps BJ, Verhoelst EM, De
Baerdemaeker JG, Decuypere EM. 2005. Dirt Detection on Brown Eggs by Means of Color Computer Vision. Poultry Science 84:1653-9.
Nair M, Jayas DS, Bulley NR. 1997. Dockage Identification in Wheat Using Machine Vision.
ASAE Annual International Meeting St, Joseph, Michigan USA:ASAE. Paper No. 973043
Nery MS, Machado AM, Campos MFM, Padua FLC, Carceroni R, Queiroz-Neto JP. 2005.
Determining The Appropiate Feature Set for Fish Classification Tasks. 18th Brazilian Symposium on Computer Graphics and Image Processing 09-12 October p173-80.
NOAA. 2007. Seafood Consumption Increases in 2006. Available from:
http://www.nmfs.noaa.gov/mediacenter/docs/06consumption_FINAL.pdf. Accessed 08-24-07
O’Sullivan MG, Byrne DV, Martens H, Gidskehaug GH, Andersen HJ, Martens M. 2003.
Evaluation of Pork Colour: Prediction of Visual Sensory Quality of Meat from Instrumental and Computer Vision Methods of Colour Analysis. Meat Science 65(2):909-18.
102
Oliveira ACM, Balaban MO. 2006. Comparison Of A Colorimeter With A Machine Vision System In Measuring Color Of Gulf Of Mexico Sturgeon Fillets. Applied Engineering in Agriculture 22(4):1-5.
Palm C. 2004. Colour Texture Classification by Integrative Co-occurrence Matrices. Pattern
Recognition 37(5):965-76. Park B, Chen YR, Nguyen M, Hwang H. 1996. Characterizing Multi-spectral Images of
Tumorous, Bruised, Skin-torn, and Wholesome Poultry Carcasses. International Transactions of the ASAE 39(5):1933-41.
Park C, Lawrence KC, Windham WR, Chen YR, Chao K. 2002. Discriminant Analysis of Dual-
Wavelength Spectral Images for Classifying Poultry Carcasses. Computers and Electronics in Agriculture 33(3):219-31.
Perez A, Pollack S. 2002. Fruit and Tree Nuts Outlook.USDA. Available from:
http://www.ers.usda.gov/publications/fts/may02/fts298.pdf. Accessed 03-11 Polder G, Van der Heijen GWAM, Young IT.2000. Hyperspectral Image Analysis for Measuring
Ripeness of Tomatoes; St. Joseph, Michigan USA:ASAE. Paper No. 003089. Quevedo R, Carlos LG, Aguilera JM, Cadoche L. 2002. Description of Food Surfaces and
Microstrutural Changes Using Fractal Image Texture Analysis. Journal of Food Engineering 53(4):361-71.
Reed TR, Hans Du Buf MH. 1993. A review of recent texture segmentation and feature
extraction techniques. CVGIP-Image Understanding 57(3):359-72. Sandusky CL, Heath JL. 1998. Sensory and Instrument-Measured Ground Chicken Meat Color.
Poultry Science 77:481-6. Scott A. 1994. Automated Continous Online Inspection, Detection and Rejection. Food
Technology Europe 1(4):86-8. Seghi RR, Hewlett ER, Kim J. 1989. Visual and Instrumental Colorimetric Assessments of Small
Color Differences on Translucent Dental Porcelain. Journal of Dental Research 68(12):1760-4.
Seida SB, Frenke EA. 1995. Unique Applications of Machine Vision for Container Inspection
and Sorting. Food Processing Automation IV. Proceedings of the FPAC Conference; St. Joseph, Michigan, USA: ASAE.
Skrede G, Trond S, Tormod N. 1989. Color Evaluation in Raw, Baked & Smoked Flesh of
Rainbow Trout (Onchorhynchus mykiss) Fed Astaxanthin or Canthaxanthin. Journal of Food Science 55(6):1574-8.
103
Strachan NJC. 1993. Length Measurements of Fish by Computer Vision. Computers and Electronics in Agriculture 8(2):93-104.
Strachan NJC, Kell L. 1995. A Potential Method for The Differentiation between Haddock Fish
Stocks by Computer Vision Using Canonical Discriminant Analysis. ICES Journal of Marine Science 52(1):145-9.
Strickland ML. 2000. Online Particle Sizing Aids in Sugar Production. Sugar Journal
62(8):14-20. Sun DW, Du CJ. 2004. Segmentation of Complex Food Images by Stick Growing and Merging
Algorithm. Journal of Food Engineering 61:17-26. Tan FJ, Morgan MT, Ludas LI, Forrest JC, Gerrard DC. 2000. Assessment of Fresh Pork Color
With Color Machine Vision. Journal of Animal Science 78(12):3078-85. Tan J. 2004. Meat Quality Evaluation by Computer Vision. Journal of Food Engineering
61:27-35. Tao Y, Heinemann PH, Varghese Z, Morrow CT, Sommer HJ. 1995. Machine Vision for Color
Inspection of Potatoes and Apples. Transactions of the ASAE 38(5):1555-61. Tuceryan M, Jain AK. 1998. Chapter 2.1 Texture Analysis. In: Chen CH, Pau LF, Wang PP,
editors. The Handbook of Pattern Recognition and Computer Vision. 2nd ed. Dartmouth, USA: World Scientific Publishing Co. p 207-48.
USDA. 2006a. Background Statistics: U.S. Beef and Cattle Industry. Available from:
http://www.ers.usda.gov/news/BSECoverage.htm. Accessed 04-01-07 USDA. 2006b. Fruit and Tree Nuts YEARBOOK Available from:
http://www.ers.usda.gov/publications/fts/2006/Yearbook/FTS2006s.txt. Accessed 03-11 USDA. 2006c. Nigeria No. 1 Market for U.S. Wheat; Potential for Other Grains and Feeds
Available from: http://www.fas.usda.gov/info/fasworldwide/ 2006/09-2006/NigeriaGrainandFeed.pdf Accessed 02-19-07
Uthu H. 2000. Application of The Feature Selection Method to Discriminate Digitised Wheat
Varieties. Journal of Food Engineering 46(3):211-6. Wallat GK, Luzuriaga DA, Balaban MO, Chapman FA. 2002. Analysis of Skin Color
Development in Live Goldfish Using a Color Machine Vision System. North American Journal of Aquaculture 64:79-84.
Wan YN, Lin CM, Chiou JF. 2000. Adaptive Classification Method for An Automatic Grain
Quality Inspection System Using Machine Vision and Neural Network. 2000 ASAE Annual International Meeting St. Joseph, Michigan, USA: ASAE. Paper No. 003094.
104
Yimyam P, Chalidabhongse T, Sirisomboon P, Boonmung S. Physical Properties Analysis of
Mango Using Computer Vision. In: The Institute of Control A, and Systems Engineers, editor; 2005 June 2-5; Gyeonggi, Korea. p 111-5.
Yoruk R, Yoruk S, Balaban MO, Marshall MR. 2004. Machine Vision Analysis of Antibrowning
Potency for Oxalic Acid: A Comparative Investigation on Banana and Apple. Journal of Food Science 69(6):281-9.
Zayas IY, Martin CR, Steele JL, Kartsevich A. 1996. Wheat Classification Using Image Analysis
and Crush Force Parameters. Transactions of the ASAE. 39(6), p 2199-204. Zheng C. 2006. Development of novel image segmentation and image feature extraction
techniques and their applications for the evaluation of shrinkage, moisture content, and texture of cooled large cooked beef joints.PhD. Dissertation. University College Dublin: National University of Ireland
. Zhu LG, Brewer MS. 1999. Relationship Between Instrumental and Visual Color In A Raw,
Fresh Beef and Chicken Model System. Journal of Muscle Foods 10:131-46. Zion B, Shklyar A, Karplus I. 1999. Sorting Fish by Computer Vision. Computers and
Electronics in Agriculture 23(3):175-87. Zuñiga-Arias, Ruben R. 2007. Variability in Quality and Management Practices in the Mango
Supply Chain from Costa Rica. Available from: http://ageconsearch.umn.edu/bitstream/123456789/27260/1/sp07zu01.pdf. Accessed 09-16-2007
105
BIOGRAPHICAL SKETCH
Jose Aparicio was born in San Pedro Sula, Honduras. He started college in Honduras and
transferred to the University of Florida in 2003 where he obtained his B.S. in dairy industry. In
2005 he gained admission to the University of Florida graduate school to work on his M.S. in the
food science program under Dr. Murat Balaban’s supervision. He completed his degree in Fall
2007.