recent developments in application of image processing techn

17
RECENT DEVELOPMENTS IN APPLICATION OF IMAGE PROCESSING TECHNIQUES FOR FOOD QUALITY EVALUATION RUKMINI MOKKAPPATY AVINASHILINGAM DEEMED UNIVERSITY, FACULTY OF ENGINEERING, COIMBATORE. ABSTRACT The paper deals with recent advances in image processing techniques or food quality evaluation. Image processing techniques have been applied increasingly for food quality evaluation in recent years. It includes charge coupled device camera, ultrasound, magnetic resonance imaging, computed tomography, and electrical tomography for image acquisition, pixel and local pre-processing approaches for image pre-processing, thresh holding based, gradient based, region based, and classification based methods for image segmentation, size, shape, color and texture features for object measurement and statistical fuzzy logic, and neural network methods for classification. The promise of image processing techniques is demonstrated and some issues which need to be resolved or investigated further to expedite the application of image processing

Upload: athmadha

Post on 20-Feb-2015

39 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Recent Developments in Application of Image Processing Techn

RECENT DEVELOPMENTS IN APPLICATION OF

IMAGE PROCESSING TECHNIQUES FOR FOOD

QUALITY EVALUATION

RUKMINI MOKKAPPATY

AVINASHILINGAM DEEMED UNIVERSITY,FACULTY OF ENGINEERING,COIMBATORE.

ABSTRACT

The paper deals with recent advances in image processing techniques or food

quality evaluation. Image processing techniques have been applied increasingly for food

quality evaluation in recent years. It includes charge coupled device camera, ultrasound,

magnetic resonance imaging, computed tomography, and electrical tomography for

image acquisition, pixel and local pre-processing approaches for image pre-processing,

thresh holding based, gradient based, region based, and classification based methods for

image segmentation, size, shape, color and texture features for object measurement and

statistical fuzzy logic, and neural network methods for classification. The promise of

image processing techniques is demonstrated and some issues which need to be

resolved or investigated further to expedite the application of image processing

technologies for food quality evaluation are also discussed.

INTRODUCTION

Quality is the main drivers of the modern food processing industry. This is the

case with items such as fruits and vegetables, fresh fish, bread, and many other

unpackaged products. In most of cases, the appearance of the product is the only direct

information received by consumers before selection and purchase.

Visual inspection, a process of interaction between product, the eyes and the

brain, can now be implemented for food quality control using digital cameras, image

Page 2: Recent Developments in Application of Image Processing Techn

analysis techniques implemented as computer software. This has resulted in computer

vision and supporting technologies being used as objective, consistent, quantitative and

non-destructive methods for evaluating and classifying foods based on their desirable

external characteristics.

Computer vision involves the acquisition, processing and analysis of images.

Image acquisition involves standardized operations in order to generate digital images

under controlled conditions. Images obtained under illumination which simulates

daylight conditions are employed to determine surface color and color patterns.

Image processing involves a series of operations that enhances the quality of an image

in order to remove defects such as geometric distortion, improper focus, background

noise, non-uniform lighting and movement. Image analysis starts by identifying and

selecting those desirable objects and separating them from the background. For

instance, one might be interested in black spots in ripened bananas, the specific color of

salmon my tomes, or light oily zones in potato chips.

A digital image is basically a two-dimensional matrix of (m x n) pixels (or

picture elements) with values from 0 to 255. Color images consist of 3 matrices, one for

each of the fundamental colors: red (R), green (G) and blue (B), while grey scale or

black and white images have only one matrix with pixels ranging from black (0) to

white (255). The matrix nature of the digital images makes them relatively easy to

process and analyze.

CCD cameras are used throughout the food industry to acquire color images for

the purposes of quality classification, detection of physical characteristics and

estimation of specific properties of food products. Illumination is important in order to

capture a good, reproducible image, since minor variations in lightning conditions can

produce considerable differences in the final result.

Image texture is usually related to the repetition of a basic pattern and it is

defined as a function of the spatial variation in pixel intensities (grey values) within an

Page 3: Recent Developments in Application of Image Processing Techn

image. Texture is one of the basic characteristics of a visible surface and plays a crucial

role in computer vision and pattern recognition. Image texture analysis has been used to

evaluate agricultural and food product quality, particularly in grading and inspection.

Applications of computer vision systems in the food industry eliminate the

subjectivity of visual inspections and add accuracy and consistency. CVSIA allows fast

identification and measurement of selected objects, high versatility for classification

into categories and multiple options for color analysis of food surfaces. It can be

designed for specific applications using readily available hardware (cameras, scanners,

computers) and commercial or free software. It is anticipated that inspection of food

products by CVSIA could be implemented on-line at reasonable costs.

QUALITY EVALUATION and sorting are important operations performed

routinely in the food industry. Regardless of the product, from fresh produce to

prepared foods, careful elimination of defective units enhances the market grade of the

product for better quality usually brings higher revenues and more important, consumer

satisfaction. In the food industry, quality evaluation is traditionally performed by human

vision of some trained operators/inspectors. However, increased demands for

Page 4: Recent Developments in Application of Image Processing Techn

consistency and speed have necessitated the introduction of computer-based

technologies such as machine vision or computer vision.

COMPUTER VISION VS HUMAN VISION

Computer vision is both better and worse than human vision. Perhaps the

biggest advantage of computer vision is its ability to be objective and consistent over

long periods. The objectivity of human vision suffers from certain inherent limitations

of our visual image perception. First of all, our visual perception is limited to 380-700

run. It also has a limit dynamic range and can discriminate fewer than 100 shades of

gray, compared to several hundred levels of gray in a vision system.

The perceived brightness of an object is not a simple function of light intensity.

The human visual system tends to overshoot or undershoot around the boundary of

regions of different intensities. This is called Mach band effect. Another phenomenon,

called simultaneous contrast, describes how an object's perceived brightness is

influenced by the surroundings. However, computer vision requires a definite time

proportional to the complexity of the decision being made. It is also very fussy about

lighting and may have difficulty in coping with reflections and with insignificant

random variations in the objects at which it is looking.

DIGITAL IMAGE PROCESSING

Essential elements of computer vision systems include the following components:

For acquisition, a high resolution (CCD) camera and a vision processor.

Board (frame grabber) for storage.

Computer disk/ tape drives for processing, computer hardware Software.

Communication, cables for display, video.

charge coupled device

Page 5: Recent Developments in Application of Image Processing Techn

A digital image is a two dimensional array' light-intensity function where the

value or amplitude of the function at a given spatial coordinate represents the light

intensity (brightness) of the image at that point. This is known as the gray level value of

the image. The gray scale values range from zero (pure black) to a maximum (pure

white) determined by the resolution of the image acquisition elements. For example, an

8-bit frame grabber would represent grayscale values in the range of 0-255 (a total of 28

= 256 gray levels). Each spatial location in the image is known as a pixel (short for

picture element). The total number of pixels is determined by the size of the two

dimensional array used (in the camera) in acquiring the image. Thus, for a given

system, a pixel is the smallest part of the digital image which can be assigned a gray

scale value. Most commonly used systems have a spatial resolution of 512 X 480 or 640

X 480.

In the vast majority of cases, however, image processing systems analyze image

features and provide quantitative information. Such quantitative information is used by

electronic sensors for decision making. Common elements of image analysis are

presented in Fig. 2. The dotted lines represent a grouping of the elements to provide a

useful framework for categorizing various processing techniques.

Page 6: Recent Developments in Application of Image Processing Techn

The groups are labeled:

low level processing

intermediate-level processing

high- level processing

LOW LEVEL PROCESSING includes image acquisition and preprocessing. Image

acquisition is the capture of an in digital form and is obviously first step in any vision

system application. Preprocessing activities include image enhancement and restoration.

It is a rather nonspecific improvement of the perceived quality of the image, including

quantitative correction of the image to compensate for degradations introduced during

image acquisition.

INTERMEDIATE LEVEL PROCESSING includes image segmentation and image

representation and description. The image segmentation operation helps break the

image into areas which correspond to physically meaningful objects. This is often

performed to isolate the object from the background information. The segmentation step

results in groups of raw pixel data which must be transformed into a form suitable for

subsequent processing.

Image representation and description deals with extracting features that result in

some quantitative information of interest or features that are basic for differentiating

one class of objects from another.

HIGH LEVEL PROCESSING involves recognition and interpretation. The

recognition step assigns a label to an object on the basis of information from its

descriptors. Interpretation is assigning meaning to recognized objects. However, for

recognition and interpretation, the knowledge and understanding of fundamental into

computer memories, enabling them to recognize and interpret images.

Page 7: Recent Developments in Application of Image Processing Techn

The back-propagation (B.P) algorithm is most commonly used for adjusting

connection weights during training. The BP networks are hierarchical, consisting of at

least three layers: an input layer, an output layer, and a hidden layer. A layer is a set of

for size, grade, and product orientation. Use of computer vision systems in analyzing

microscopic images is also gaining popularity for internal product quality assessment

and as a means for studying effects of variations in composition and processing

parameters. Another class of applications that holds much promise for the food industry

is automatic process control. This includes guidance operations to control orientation,

sorting, packing, and delivery systems. Object shape is one of the most commonly

applied quality criteria. Most foods, such as grains, fruits, vegetables, nuts, and other

prepared snacks, have certain shape features that signify their overall quality. Thus,

damage to these foods usually results in a change in object profile and shape. Therefore,

shape inspection is widely needed in the food industry. Identifier broken crackers from

good ones is used below to illustrate the application of a computer vision system in

conjunction with artificial neural nets. The basic idea is that if shape features of a

reference cracker can be obtained, the good and broken crackers can be identified by

comparing the corresponding shape features of the cracker under inspection to those of

the reference cracker. The image of the reference cracker is the average of several good

crackers. Distinguishing shape features include area, aspect ratio (ratio of length to

width), curvature and continuity of the edge, and radius (from a center location to the

edge). The cracker surface area is readily obtained by counting the pixels representing

the image surface. Several commercial image processing software programs can readily

perform pixel counting and provide the area.

For non homogeneous shaped objects such as crackers, length can be defined as

the longest dimension through a certain location. The width is then the longest

dimension perpendicular to the length. Unlike area and aspect ratio, the curvature,

continuity, and radius are specific to a point and/or to its neighboring points along the

edge. Therefore, 32 equal angular points along the image were chosen after positioning

the cracker image in a reference location, orientation, and scale. A 3-layer BP neural net

classifier was used with five input nodes, five hidden nodes, and two output nodes. The

Page 8: Recent Developments in Application of Image Processing Techn

inputs comprised the five shape indices, and the outputs were trained with three sets of

crackers, each set containing 24 good crackers and 24 broken ones. The crackers were

purchased from store, and the damage was manually induced by breaking the crackers

to various extents. The extent of damage was intentionally small to test the system in a

demanding application. BP classifier always produced better results than the Bayes

classifier. In decreasing order of classification error, the five shape features can be

ordered as continuity, radius, curvature, aspect ratio, and area, respectively. It was

surprising to note that area was the least successful feature in identifying the broken

crackers. This result is caused by minor variations in the size of good crackers which

arise during cutting and baking. The advantage of using multiple shape features via

neural nets is thus clear. A similar example is detecting dam, led almonds. Digital

images of group almonds randomly selected from a commercial package. Algorithm

similar to cracker classification can be successfully applied for almonds as well.

Page 9: Recent Developments in Application of Image Processing Techn

APPLICATION IN QUALITY ASSESSMENT OF FRUITS

Computer vision has been used for quality inspection of fruits. Quality inspections

of fruits have two different objectives: quality evaluation and defect finding.

The study on apples using computer can reflect the progress of computer vision

technology for fruit inspection. Computer vision can implement the tasks of shape

classification, defects detection, and quality grading and variety classification. Color,

shape, texture and size are the main factors corresponding to quality inspection by

computer vision. And the results obtained by monochrome algorithms are normally

dissatisfactory. RGB (Red, Green and Blue) color system is the most popular color

model. When color is presented with red, green and blue, the amount of information is

tripled. This algorithm is effective in representing minute structures with high quality. So

it is suitable for describing the similar-spherical haps, such as apple and potato.

VEGETABLE INSPECTION

Tomato is one of the most popular vegetables in the world. Nielsen et al. (1998)

developed a method to use the attributes of size, color, shape and abnormalities, obtained

from tomato images, to correlate with the inner.

APPLICATIONS IN OTHER FOOD PRODUCTS

Successfully extracted pizza topping percentage and distribution from the pizza images.

For the complex visual features and wide varieties of pizzas, any single segmentation

algorithm such as threshold, edge-based segmentation or region-based segmentation,

could not achieve satisfactory results.

Page 10: Recent Developments in Application of Image Processing Techn

Segmented image of a pizza

CONCLUSION

Shape and microstructure are related to physical damage and physical properties

of food. Along with the demands of food quality, the automated food shape inspection

and the quantitative food microstructure evaluation has become increasingly important.

In this study, some image processing and computer vision techniques were developed and

applied for food shape and microstructure evaluation. Food shape inspection using

computer vision has been studied for a long time. However, the accuracy and speed are

still two major problems. To improve the accuracy of the shape feature extraction, a

multi-index active model-based (MAM) feature extractor was developed. Experiments

showed that generally the MAM feature extractor was more accurate (5% or more in

correct classification rate) than the single-index, statistical, or non-model-based methods.

In the classification stage, the back propagation neural network (BP net) was initially

applied for the food shape classification. Then, some minimum indeterminate zone (MIZ)

classifiers were developed to speed up the training. The above techniques were tested by

inspecting grain kernels, nuts and crackers. The experiments showed that the correct

classification rates obtained using the Maxim MIZ classifier were comparable with those

obtained using the BP net. However, when the Maxim Mize classifier was used, the

training times reduced by two to four orders of magnitude. In the second part of the

thesis, microstructure of cheese and milk gel was evaluated. Some two-dimensional (2-D)

image processing techniques including curve and image surface complexities were

developed for analyzing the micro structural differences of milk gel images from a

Page 11: Recent Developments in Application of Image Processing Techn

transmission electron microscope. In addition, some three-dimensional (3-D) image

reconstruction and analysis techniques were developed for analyzing cheese microscopic

images obtained by co focal laser scanning microscope (CLSM). The 3-D image analysis,

in conjunction with 2-D techniques, provided very useful information for evaluating the

cheese microstructure.