Acta Psychologica xxx (2011) xxx–xxx

ACTPSY-01673; No of Pages 10 July 08, 2011; Model: Gulliver 5

Contents lists available at ScienceDirect

Acta Psychologica

j ourna l homepage: www.e lsev ie r.com/ locate /actpsy

The role of color information on object recognition: A review and meta-analysis

Inês Bramão a,⁎, Alexandra Reis a, Karl Magnus Petersson a,b,c,d, Luís Faísca a

a Cognitive Neuroscience Research Group, Institute of Biotechnology & Bioengineering/CBME, Universidade do Algarve, Faro, Portugalb Cognitive Neurophysiology Research Group, Stockholm Brain Institute, Karolinska Institutet, Stockholm, Swedenc Max Planck Institute for Psycholinguistics, Nijmegen, The Netherlandsd Donders Institute for Brain, Cognition, and Behaviour, Radboud University Nijmegen, The Netherlands

⁎ Corresponding author at: Departamento de PsicoloTel.: +351 289800 900x7660; fax: +351 289819 403.

E-mail address: mibramao@ualg.pt (I. Bramão).

0001-6918/$ – see front matter © 2011 Elsevier B.V. Aldoi:10.1016/j.actpsy.2011.06.010

Please cite this article as: Bramão, I., et al., T(2011), doi:10.1016/j.actpsy.2011.06.010

a b s t r a c t

a r t i c l e i n f o

Article history:Received 18 November 2010Received in revised form 21 June 2011Accepted 25 June 2011Available online xxxx

PsycINFO classification:2300 (Human Experimental Psychology)2323 (Visual Perception)

Keywords:Meta-analysisLiterature reviewColor informationObject recognitionColor diagnosticityObject properties

In this study, we systematically review the scientific literature on the effect of color on object recognition.Thirty-five independent experiments, comprising 1535 participants, were included in a meta-analysis. Wefound a moderate effect of color on object recognition (d=0.28). Specific effects of moderator variables wereanalyzed and we found that color diagnosticity is the factor with the greatest moderator effect on theinfluence of color in object recognition; studies using color diagnostic objects showed a significant color effect(d=0.43), whereas a marginal color effect was found in studies that used non-color diagnostic objects(d=0.18). The present study did not permit the drawing of specific conclusions about the moderator effect ofthe object recognition task; while the meta-analytic review showed that color information improves objectrecognition mainly in studies using naming tasks (d=0.36), the literature review revealed a large body ofevidence showing positive effects of color information on object recognition in studies using a large variety ofvisual recognition tasks. We also found that color is important for the ability to recognize artifacts and naturalobjects, to recognize objects presented as types (line-drawings) or as tokens (photographs), and to recognizeobjects that are presented without surface details, such as texture or shadow. Taken together, the results ofthe meta-analysis strongly support the contention that color plays a role in object recognition. This suggeststhat the role of color should be taken into account in models of visual object recognition.

gia, Faculdade de Ciências Humanas e Sociais, Campus

l rights reserved.

he role of color information on object recogni

© 2011 Elsevier B.V. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 01.1. Color diagnosticity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 01.2. Semantic category . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 01.3. Recognition task. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 01.4. Stimulus type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 01.5. Surface detail . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 01.6. Snodgrass and Vanderwart set (1980) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0

2. Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 02.1. Study selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 02.2. Data extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0

2.2.1. Color diagnosticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 02.2.2. Semantic category . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 02.2.3. Object recognition task . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 02.2.4. Stimulus type. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 02.2.5. Surface details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 02.2.6. Snodgrass and Vanderwart set (1980) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0

2.3. Effect size estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 03. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0

3.1. Overall results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 03.2. Color diagnosticity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0

de Gambelas, Universidade do Algarve, 8005–139 Faro, Portugal.

tion: A review and meta-analysis, Acta Psychologica

2 I. Bramão et al. / Acta Psychologica xxx (2011) xxx–xxx

3.3. Semantic category . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 03.4. Recognition task . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 03.5. Stimulus type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 03.6. Surface details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 03.7. Snodgrass and Vanderwart set (1980). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0

4. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 05. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0Acknowledgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0

1. Introduction

Traditionally, theories of object recognition suggest that objectsare recognized based on shape information, largely ignoring thepotential role of color information (Biederman, 1987; Marr &Nishihara, 1978). For example, the recognition-by-componentsmodel, proposed by Biederman (1987), hypothesizes that objectsare represented in terms of geons, basic geometric building blocks.This model assumes that to identify an object, the perceptual systemcomputes a structural description – it determines the geons of theobject and the relations among them – and in turn, this descriptionprovides access to function and meaning, as well as the object name.More importantly, neither geons nor the relations among them areassociated with color information or color knowledge. More recently,a large body of behavioral, functional neuroimaging, and neurophys-iological evidence suggests that color information contributes toobject recognition, and for that reason, the role of color should beintegrated in object recognition models (for a review, see Tanaka,Weiskopf, & Williams, 2001). However, although color information isnow accepted to contribute to object recognition, the objectproperties and the viewing conditions that might benefit from colorinformation are not well-understood. In this review, and accompa-nying meta-analysis, we discuss the behavioral literature on the effectof color on object recognition as well as draw conclusions regardingthe moderator role of several variables that are typically manipulatedin studies that examine the influence of color on object recognition.

1.1. Color diagnosticity

Color diagnosticity is probably the most investigated property instudies exploring the role of color information in object recognition.Color diagnosticity is defined as the degree to which a particularobject is associated with a specific color. For example, a strawberry – acolor diagnostic object – is clearly associated with the red color,whereas a comb – a non-color diagnostic object – is not stronglyassociated with any particular color. It has been proposed that colorinformation is more important for the recognition of color diagnosticobjects. For example, Tanaka and Presnell (1999) found that coloredversions of color diagnostic objects were recognized faster thanuncolored versions, while non-color diagnostic objects were recog-nized equally fast in color and in black and white. Similar results werereported by Nagai and Yokosawa (2003). However, other studies havenot replicated these findings and provide evidence that colorinformation, independent of diagnosticity status, improves recogni-tion (Bramão, Faísca, Petersson, & Reis, 2010; Rossion & Pourtois,2004; Uttl, Graf, & Santacruz, 2006; see also Biederman & Ju, 1988 andWurm, Legge, Isenberg, & Luebker, 1993).

A possible explanation for the discrepancy is that differentmethods have been used to determine the color diagnosticity of anobject. Tanaka and Presnell (1999) used a very strict method toclassify the color diagnosticity of an object. Color diagnosticity wasdetermined based on (1) feature listing (subjects were instructed tolist perceptual properties of the object) and (2) typicality judgmenttask (subjects were asked about the typical color of the object). An

Please cite this article as: Bramão, I., et al., The role of color information(2011), doi:10.1016/j.actpsy.2011.06.010

object was classified as a high color diagnostic object if a color waslisted first in the feature list and if at least 80% of the subjects agreedon the typical color of the object. A similar method was used by Nagaiand Yokosawa (2003). Less strict methods were used in the otherstudies. For example, in Biederman and Ju (1988), three independentraters determined whether a color was diagnostic or not for a givenobject. Wurm and colleagues provided subjects with a color name,and participants rated the relative symptomaticity of the color for agiven object. An object was classified as high in color diagnosticity if acolor was highly symptomatic of the object and not symptomatic ofother objects (Wurm et al., 1993). Rossion and Pourtois (2004) askeda group of participants to rate objects on a 5-point scale (1=a colorwas not diagnostic for the object; 5=high diagnosticity). A similarmethodwas used by Uttl and colleagues (Uttl et al., 2006) and Bramãoand colleagues (Bramão, Faísca, Petersson, & Reis, 2010; Bramão,Inácio, Faísca, Reis, & Petersson, 2011).

It is possible that when a stricter method is used to define colordiagnosticity, such as the one used by Tanaka and Presnell (1999),color information facilitate the recognition of color diagnostic objectsonly. When less strict criteria are used, objects not so stronglyassociated with a specific color might be classified as diagnostic. Infact, we observed that items classified as low color diagnostic objectsby Tanaka and Presnell (1999) were considered as color diagnosticobjects by Rossion and Pourtois (2004). For example, nail and forkwere considered as low color diagnostic objects by Tanaka andPresnell (1999); however, Rossion and Pourtois (2004) found colordiagnostic rates of 4.45 and 4.09, respectively, on their 5 point scale.

1.2. Semantic category

The role of color in the recognition of objects from differentsemantic categories has been addressed in the literature. Price andHumphreys (1989) found that naming of objects from naturalcategories was facilitated by color. Because objects from naturalcategories tend to be structurally more similar than artifacts,competition within the object recognition system is greater fornatural objects than artifacts, and color information serves animportant role in resolving this competition. Moreover, Wurm et al.showed that prototypical images exhibit a smaller color advantagecompared to non-prototypical images (Wurm et al., 1993). Theseobservations suggest that color plays an important role in objectrecognition when shape is not diagnostic or typical. In a recent study,Laws and Hunter (2006) examined the role of color and blurring intwo naming experiments across natural and artifact categories. Whenthe objects were presented in a non-blurred format there was no coloradvantage in naming accuracy for either category (the error rateswere low, so a ceiling effect cannot be ruled out). However, when theobjects were presented in a blurred format, a color advantage wasfound for the natural categories, but not for artifacts. Laws and Hunter(2006) argued that the blurring increase the level of visual crowdingand that color information is used to segment the shape componentsof the natural objects. Alternatively, the observed color advantage fornatural objects might be related to the fact that these objects aretypically more strongly associated with a specific color, or a small

on object recognition: A review and meta-analysis, Acta Psychologica

3I. Bramão et al. / Acta Psychologica xxx (2011) xxx–xxx

range of colors, and therefore, color tends to be more diagnostic fornatural objects compared to artifacts. This interaction betweencategory and color diagnosticity was addressed by Nagai andYokosawa (2003), who reported a color advantage for high colordiagnostic objects regardless of their category. Other studies havereported similar findings (Bramão, Faísca, Forkstam, Reis, & Petersson,2010; Rossion & Pourtois, 2004; Uttl et al., 2006).

1.3. Recognition task

The studies reviewed here typically used naming, object-nameverification, and semantic classification (natural versus artifact) toevaluate the role of color in object recognition. Different objectrecognition tasks impose different cognitive demands (Humphreys,Price, & Riddoch, 1999). To name an object, its semantic representa-tion must be activated and its name retrieved (or activated). Incontrast, only the semantic representation needs to be activated in asemantic classification task. A dissociation of color effects in these twotypes of tasks provides information on which visual recognition stagecolor information affects in the object recognition process. To matchan object with a previous presented object name, subjects needed toactivated semantic and object name representations. Biederman andJu (1988) did not find any advantage of color in semantic verification.Nevertheless, they found a significant advantage of color in one oftheir object naming experiments (unmasked condition). This advan-tage was not found in the masked condition. Davidoff and colleaguesalso failed to find any color effect in semantic classification butreported an advantage of color in object naming (Davidoff &Ostergaard, 1988; Ostergaard & Davidoff, 1985).

The finding that color information improves object naming to agreater degree than semantic categorization led some researchers topropose that color plays a role in the latter stages of visual processing(i.e., after semantic access), providing an associative link between theshape representations of an object and its name (Davidoff &Ostergaard, 1988; Davidoff, Walsh, & Wagemans, 1997; Tanaka etal., 2001). However, a number of recent studies have not replicatedthese findings and have instead reported a recognition advantagerelated to color information for both naming and semantic verificationof objects (see, for example, Therriault, Yaxley, & Zwaan, 2009) as wellas visual scenes (see, for example, Oliva & Schyns, 2000). Thesefindings suggest that the role of color is not restricted to the access ofthe name. In line with this hypothesis, several studies have reportedcolor effects in the early stages of visual processing (Gegenfurtner &Rieger, 2000; Wurm et al., 1993). For example, Wurm et al. found thatcolor improved object identification irrespective of color diagnosti-city, and suggested that this provides evidence for an early low-levelsensory processing contribution (Wurm et al., 1993). In addition,others have argued that color is represented in a structuralrepresentation system at a perceptual level (Price & Humphreys,1989) and/or at a semantic level, where conceptual knowledge ofprototypical object color is stored and provides an associative linkbetween object shape and object name representations (Davidoffet al., 1997). In support of this idea, there is evidence suggesting thatstored knowledge of object color also plays a role in object iden-tification (Joseph, 1997; Joseph & Proffitt, 1996; Mapelli & Behrmann,1997).

1.4. Stimulus type

Most of the studies that evaluate the role of color in objectrecognition have compared black-and-white and color line-drawings(see, for example, Vernon & Lloyd-Jones, 2003) or black-and-whiteand color photographs (see, for example, Lloyd-Jones & Nakabayashi,2009). Recently, Uttl and colleagues (Uttl et al., 2006) suggested that aline-drawing of an object is typically viewed as a representative of anobject class – a type – while photographs are viewed as individual

Please cite this article as: Bramão, I., et al., The role of color information(2011), doi:10.1016/j.actpsy.2011.06.010

objects — a token. The recognition of types and tokens may recruitdifferent perceptual and semantic processes and for that reason colorinformation might make a different contribution to the recognition ofline-drawings compared to the recognition of photographs. Onlythree studies, investigating the role of color information in objectrecognition, have explored line-drawings and photographs of thesame objects. Two studies reported a similar color advantage for bothstimulus types (Bramão et al., 2011; Price & Humphreys, 1989). Thethird study compared the performance of illiterate and literate elderlysubjects in a naming task. The results showed that the illiteratesubjects benefitted from color information in line-drawings andphotographs, the literate subjects only benefitted from color in line-drawings and not in photographs (Reis, Faísca, Ingvar, & Petersson,2006). In a recent study, Adlington, Laws, and Gale (2009b)investigated the naming in a group of patients with Alzheimer'sdisease (AD) and elderly controls using color and monochromaticphotographs as well as black and white line drawings derived fromthe photographs. The results showed that, while the naming accuracyof the AD patients did not improve, it did improve in the control groupwhen color or surface detail (photographs) was added to the stimulusitems.

1.5. Surface detail

Color is typically displayed together with other surface properties,including surface texture and shadow details. This is the case for moststudies reviewed here and none of these studies investigated the coloreffect with and without concomitant presentation of other surfaceproperties. Thismakes it difficult to distinguish the color effect from theeffect of other surface details. However, four independent experimentsreported a color effect using images without other surface propertiesthan color (Joseph, 1997; Moore & Price, 1999; Vernon & Lloyd-Jones,2003). This suggests that color alone is a property that can improveobject recognition.

1.6. Snodgrass and Vanderwart set (1980)

The Snodgrass and Vanderwart (1980) set of images is one of themost used in cognitive experimental research. Snodgrass andVanderwart (1980) presented a normative picture set of 260 linedrawings of common objects from different semantic categories,together with normative data for familiarity, visual complexity andname agreement for the English language. Subsequently, this set ofobjects have been standardized in different languages, includingFrench (Alario & Ferrand, 1999; Bonin, Peereman, Malardier, Méot, &Chalard, 2003), Italian (Dell'Acqua, Lotto, & Job, 2000), Spanish(Cuetos, Ellis, & Alvarez, 1999; Sanfeliu & Fernandez, 1996) andPortuguese (Ventura, 2003). More recently, Rossion and Pourtois(2004) modified the Snodgrass and Vanderwart line-drawings byadding texture and shadow details. In order to generalize theconclusions concerning for example color effects in object recognition,it is of interest to explore these effects using also other sets of objects.

2. Methods

2.1. Study selection

The studies included in the meta-analysis were identified andselect by searching the PubMed and PsycINFO databases, using thesearch terms “surface detail”, “surface information”, “colo(u)rdiagnosticity”, “colo(u)r AND object recognition”, and “colo(u)r ANDobject identification” during March and April 2010. This procedureidentified 93 articles. The title and abstract of the initial set of articleswere screened for potential inclusion, leaving 34 studies. For inclusionin the meta-analysis, a study had to meet the following criteria:1) report response time data from object recognition tasks; 2) present

on object recognition: A review and meta-analysis, Acta Psychologica

4 I. Bramão et al. / Acta Psychologica xxx (2011) xxx–xxx

the stimuli in typical color and in a black and white or gray-scaleversion; 3) use healthy participants between 18 and 60 years of age;4) contain information that allowed the computation of effect sizes.Thus, studies not reporting response times (Laws & Hunter, 2006; Uttlet al., 2006), that used visual scenes (Gegenfurtner & Rieger, 2000;Oliva & Schyns, 2000), or only typical or atypical colored objects(Joseph & Proffitt, 1996; Naor-Raz & Tarr, 2003) were excluded fromthemeta-analysis. Also, studies presenting data from elderly (Boucart,Despretz, Hladiuk, & Desmettre, 2008; Reis et al., 2006), brain lesioned(Mapelli & Behrmann, 1997) or participants with non-normal vision(Boucart et al., 2008) were not included in the meta-analysis. Onestudy was excluded because it did not provide sufficient information tocompute the effect size (Nagai & Yokosawa, 2003). In general,participants perform close to ceiling on object recognition in terms ofaccuracy in most of these studies, and for that reason we did notconsider the accuracy data in the meta-analysis. The average error ratefor the studies included in the meta-analysis is less than 5%. From theinitial 34 studies, only 10 met all of the inclusion criteria. Thus, some ofthe studies included in the literature review of the Introduction sectionwere not included in the meta-analysis. In addition, the bibliographiesfrom the 10 papers identified as outlined were inspected, and eightadditional relevant references were identified. Selected studies wererestricted to those appearing in English language journals, with theexception of one non-published study (Faísca et al., 2004). The non-published study was conducted in our lab and intended to establishnormative data for the Portuguese population related to namingresponse times for a set of 70 object representations. In this study, theobjects were presented both as line-drawings and as photographs. Theline-drawings, selected from the original Snodgrass and Vanderwart(1980) set, were presented as contours (without surface details) and ina grey-scale and colored version selected from the set generated byRossion and Pourtois (2004). The photographswere also presented in acolor and in a black and white version thatmatched as far as possible interms of color, size, shape and orientation to the line-drawings. Insummary, the 18 resultant articles and the non-published study yielded35 independent experiments in which object recognition performance,evaluated in terms of response time, was tested with typically coloredand black and white versions (see Fig. 1).

2.2. Data extraction

For each study, we extracted information about the stimuluscharacteristics: color diagnosticity status (diagnostic versus non-

Fig. 1. Flow chart of studies considered and finall

Please cite this article as: Bramão, I., et al., The role of color information(2011), doi:10.1016/j.actpsy.2011.06.010

diagnostic color), semantic category (natural object versus artifact),stimulus type (line-drawings versus photographs), surface details(present versus absent) and stimulus set (Snodgrass and Vanderwart(1980) set versus other sets). We also identified for each study therecognition task used. Finally, for each study we checked whetherlow-level visual properties of the images (for instance, luminance andcontrast) were controlled between the black-and-white and the colorconditions (see Table 1). This information is important in assessingthe moderator effect on the role of color in object recognition of thefollowing variables: (1) color diagnosticity; (2) semantic category;(3) type of recognition task; (4) stimuli type (line-drawing versusphotographs); (5) presence of surface details; and (6) if theexperimental stimuli belong to the Snodgrass and Vanderwart(1980) set.

2.2.1. Color diagnosticityNone of the studies in the meta-analysis tested the effects of color

exclusively for non-diagnostic objects. To maximize the likelihood ofmeeting the methodological assumption that effect size estimatestaken from individual studies are independent of each other,whenever a study presented both the information for color andnon-color diagnostic objects, we only selected the information aboutthe non-color diagnostic objects. Thus, the moderator effect of colordiagnosticity was assessed by comparing the effect size estimatedfrom studies that tested the color effect only in color diagnosticobjects with the effect size from studies that tested color effects onboth color and non-color diagnostic objects but using only theinformation about the non-color diagnostic objects.

2.2.2. Semantic categoryOnly one of the studies in themeta-analysis tested the color effects

exclusively in artifacts (Brodie, Wallace, & Sharrat, 1991). To test thesemantic category influence on the color effect, while maximizing thelikelihood of meeting the methodological assumption that effect sizevalues came from independent studies, we used a similar procedure tothe one adopted to evaluate the color diagnosticity effects: whenevera study presented both the information for natural objects andartifacts, we only selected the information about the objects belongingto the artifact category. Consequently, the effect of the semanticcategory was assessed by comparing the effect size estimated fromthe studies that tested the color effect only in natural objects with theeffect size from studies that tested both artifacts and natural objectsbut using only the information about the artifacts.

y selected for inclusion in the meta-analysis.

on object recognition: A review and meta-analysis, Acta Psychologica

Table 1Characteristics of the studies included in the meta-analysis.

Study d N ColorDiagnosticity

Semanticcategory

Task Colorstimuli

B&Wstimuli

Colordetails?

B&Wdetails?

S&V? Visualproperties?

Biederman and Ju (1988) — Exp. 1 −0.18 30 ND/DI AO/NA naming PH LD Yes No No Not reportedBiederman and Ju (1988) — Exp. 2 0.18 30 ND/DI AO/NA naming PH LD Yes No No Not reportedBiederman and Ju (1988) — Exp. 3 0.18 30 ND/DI AO/NA naming PH LD Yes No No Not reportedBiederman and Ju (1988) — Exp. 4 −0.10 96 ND/DI AO/NA VT PH LD Yes No No Not reportedBiederman and Ju (1988) — Exp. 5 0.10 96 ND/DI AO/NA VT PH LD Yes No No Not reportedBramão, Faísca, Forkstam, et al. (2010) 0.55 20 ND/DI AO naming LD LD Yes Yes Yes ControlledBramão, Faísca, Petersson, and Reis (2010) 0.12 28 ND AO VT PH PH Yes Yes No ControlledBramão et al. (2011) −0.02 144 ND AO/NA SCT PH/LD PH/LD Yes Yes No ControlledBrodie et al. (1991) – Exp. 3 0.00 18 ND/DI AO naming PH PH Yes Yes No ControlledBrodie et al. (1991) – Exp. 4 0.55 15 ND/DI AO OVT PH PH Yes Yes No ControlledChao and Martin (1999) −0.11 12 — — naming — — — — — Not reportedDavidoff and Ostergaard (1988) — Exp. 1 0.32 32 ND/DI AO/NA SCT LD LD Yes No Yes ControlledDavidoff and Ostergaard (1988) — Exp. 2 0.83 16 ND/DI AO/NA SCT LD LD Yes No Yes ControlledFaísca et al. (2004) 0.08 60 ND AO naming PH/LD PH/LD Yes Yes Yes/no ControlledGale, Laws, and Foley (2006) — Exp. 4 0.39 32 ND/DI AO naming LD LD Yes No Yes ControlledHocking and Price (2008) 0.33 15 ND/DI AO/NA naming PH PH Yes Yes No Not reportedHumphreys et al. (1994) — Exp 2 0.51 37 ND AO naming PH PH Yes Yes No ControlledHumphreys et al. (1994) — Exp 3 1.64 30 DI NO naming PH PH Yes Yes No ControlledJoseph (1997) 0.77 23 DI NO VT LD LD No No No Not reportedLloyd-Jones and Nakabayashi (2009) 0.59 21 DI AO/NA naming PH PH Yes Yes No ControlledMoore and Price (1999) — Exp. 1 0.24 8 ND AO naming LD LD No No No Not reportedOstergaard and Davidoff (1985) — Exp. 1 0.65 45 DI NO naming PH PH Yes Yes No ControlledOstergaard and Davidoff (1985) — Exp. 2 0.36 75 DI NO naming PH PH Yes Yes No ControlledOstergaard and Davidoff (1985) — Exp. 3 0.38 32 DI NO naming PH PH Yes Yes No ControlledPrice and Humphreys (1989) — Exp 1 0.13 50 DI NO SCT PH/LD PH/LD Yes/no Yes/no No ControlledPrice and Humphreys (1989) — Exp 2 0.31 25 DI NO SCT PH/LD PH/LD Yes/no Yes/no No ControlledRyan, Hemmes, and Brown (2003) — Exp. 2 0.05 32 ND/DI AO/NA naming LD LD Yes Yes No Not reportedRossion and Pourtois (2004) 0.71 180 ND AO naming LD LD Yes Yes Yes ControlledTanaka and Presnell (1999) — Exp. 2 0.03 45 ND AO/NA VT LD LD Yes Yes No ControlledTanaka and Presnell (1999) — Exp. 3 0.10 36 ND AO/NA naming LD LD Yes Yes No ControlledTanaka and Presnell (1999) — Exp. 4b 0.04 30 ND AO/NA VT LD LD Yes Yes No controlledTherriault et al. (2009) — Exp. 1 0.18 84 DI AO/NA VT PH PH Yes Yes No Not reportedVernon and Lloyd-Jones (2003) — Exp. 1a 0.40 30 DI NO naming LD LD No No Yes ControlledVernon and Lloyd-Jones (2003) — Exp. 1b 0.35 30 DI NO naming LD LD No No Yes ControlledWurm et al. (1993) — Exp. 2 0.41 48 DI NO naming PH PH Yes Yes No Controlled

Effect sizes (d) for each study, the total number of subjects these effect sizes were based upon (N). DI — color diagnostic objects, ND — non-color diagnostic object. NO — naturalobjects, AO— artifacts objects. VT— verification task, OVT— object verification task, SCT— semantic classification task. LD— line-drawings, PH— photographs. S&V?— Did the studyuse the Snodgrass and Vanderwart set (1980)? Visual properties? — Did the study control the low level visual properties?

5I. Bramão et al. / Acta Psychologica xxx (2011) xxx–xxx

2.2.3. Object recognition taskOnly one experiment included in the meta-analysis tested the

color effects in object recognition using exclusively a semanticclassification task (where subjects had to decide if a present objectwas from a natural or an artifact semantic category) (Price &Humphreys, 1989). Five experiments tested the color effects usingexclusively verification tasks (where subjects had to match a previouspresented name/object with a object/name) (Biederman & Ju, 1988;Bramão, Faísca, Petersson, & Reis, 2010; Tanaka & Presnell, 1999). Totest the moderator effect of the object recognition task, whileassuming that the effect size values came from independent studies,we employed a similar procedure to the one adopted to evaluate thecolor diagnosticity effects: whenever a study presented both theinformation for semantic classification task or verification tasktogether with other visual tasks, we only selected the informationabout the semantic classification or the verification task. As a result,the moderator role of the recognition task was evaluated bycontrasting the effect size estimated from studies that assessedcolor effects in naming tasks with the effect size from studies thatused semantic classification tasks and from studies that usedverification tasks (object-name or name-object verifications).

2.2.4. Stimulus typeThe effect of the stimulus type was assessed by comparing the

effect size estimated from studies that compared color and black andwhite photographs with effect size from studies that compared colorand black and white line-drawings.

Please cite this article as: Bramão, I., et al., The role of color information(2011), doi:10.1016/j.actpsy.2011.06.010

2.2.5. Surface detailsThe effect of surface details was evaluated by comparing the effect

size from studies that used stimuli with surface details both in thecolored and in the black and white version of the stimuli with theeffect size estimated from studies that used colored and black andwhite objects without surface details. The surface details consideredin the meta-analysis were texture and shadow details.

2.2.6. Snodgrass and Vanderwart set (1980)The object set effect was evaluated by comparing the effect size

estimated from studies that used the Snodgrass and Vanderwart(1980) set or its colored version from Rossion and Pourtois (2004) setwith the effect size estimated from studies that used another objectset. The studies using photographs were not included in thiscomparison.

2.3. Effect size estimates

Data were analyzed with the Comprehensive Meta-Analysissoftware v.2.2. For each color versus black and white comparison,we calculated Cohen's d to estimate the magnitude of the color effecton the response time data. When means and standard deviation werenot provided, d values were estimated from the reported t or Fstatistics. A positive d value indicates a color effect. By convention, anabsolute effect size b0.2 is considered small, an absolute value N0.2and b0.6 is moderate, and an absolute value N0.6 is considered a largeeffect. For each meta-analysis, we calculated the 95% confidence

on object recognition: A review and meta-analysis, Acta Psychologica

6 I. Bramão et al. / Acta Psychologica xxx (2011) xxx–xxx

interval (CI), statistical significance (p), within-group heterogeneity(Qwithin), and the percentage of variation across studies due toheterogeneity rather than sampling error (I2). For additionalclarification of differences between effect size estimates, we pro-ceededwith a subgroup analysis to test themoderator variable effects,with mixed-effects between-group heterogeneity (Qbetween). Studiesvaried according to sample size (range 8–180); this creates a risk thata small, outlying sample exerts disproportionate influence on themean effect. To minimize this risk, we weighted the effect sizeestimates by sample size (Rosenthal, 1991). When individual studiesincluded multiple independent experiments, separate effect sizeswere calculated for each experiment. When studies presentedinformation sufficient to derive more than one effect size estimatefor an individual experiment, effect size estimates were in some cases(for example, in the overall analysis) aggregated using the arithmeticmean. This aggregation prior to meta-analytic integration is necessaryto avoid the over-representation of multi-experiment studies in theoverall analyses (Rosenthal, 1991).

3. Results

3.1. Overall results

An overall effect size was calculated that incorporates all 35effect sizes, comprising a total of 1535 subjects. The results showeda significant moderate color information effect on object recognition(d=0.28; 95% CI=0.19–0.38; pb0.01). The heterogeneity test wassignificant (Qwithin=88.9; pb0.01; I2=61.7%), suggesting that morethan two-thirds of the observed variance was not accounted for bysampling error. This implied that further meta-analytic subdivisionof the overall sample was warranted. A systematic analysis oftheoretically meaningful, a priori selected, moderator variables wastherefore conducted on six subsets of the overall pool of studies.The results of the separated meta-analysis are given in Fig. 2 andTable 2.

Fig. 2.Mean effect size (d) and 95% confidence intervals for the 13meta-analyses conducted.left side. Labels to the right side of the figure indicate the number of independent effect sizesthese effect sizes were based upon (Ns).

Please cite this article as: Bramão, I., et al., The role of color information(2011), doi:10.1016/j.actpsy.2011.06.010

3.2. Color diagnosticity

Color diagnosticity was assessed by comparing studies where thecolor effect was evaluated using color diagnostic objects with studiesusing non-color diagnostic objects. This comparison was significant(Qbetween=4.28, p=0.04), with larger effects in studies where thecolor effect was assessed using color diagnostic objects. The studiesusing non-color diagnostic objects showed a marginally significantcolor effect (d=0.18, p=0.06), whereas studies using color diagnos-tic objects showed a moderate effect (d=0.43, pb0.01).

3.3. Semantic category

The color effect in studies that used natural object categories wascompared with studies that used artifacts. No difference in the coloreffect was found (Qbetween=0.59, p=0.44). Studies that used naturalobjects and artifacts showed a similar color advantage (naturalobjects d=0.45, pb0.01; artifacts d=0.36, pb0.01).

3.4. Recognition task

To examine the moderator effect of recognition task, we comparedstudies that used naming with those that used semantic classificationand verification tasks. The comparison was significant (Qbetween=6.46, p=0.04); naming showed a moderate color effect (d=0.36,pb0.01), while verification tasks did not show a significant color effect(d=0.11, p=0.15). Studies that used semantic classification showeda marginally significant color effect (d=0.23, p=0.06).

3.5. Stimulus type

The color effect in studies that used line-drawings was comparedwith studies that used photographs. No difference in the color effectwas found (Qbetween=0.001, p=0.97); both showed amoderate effect(line-drawings d=0.35, pb0.01; photographs d=0.34, pb0.01).

Themoderator variables tested in specific meta-analytic comparisons are labeled on the(experiments) that contributed to each meta-analysis (NE), and the number of subjects

on object recognition: A review and meta-analysis, Acta Psychologica

Table 2Effect sizes and comparisons across the subgroups.

Effect size and 95% confidence interval Heterogeneity

Ne Ns d Lower limit Upper limit Z-value p value Qwithin p value I2 Qbetween p value

Overall 35 1535 0.28 0.19 0.38 5.9 0.00 88.87 0.00 61.74Color diagnosticity

Color diagnostic objects 12 493 0.43 0.28 0.58 5.73 0.00 24.12 0.01 54.40Non-color diagnostic objects 9 568 0.18 −0.01 0.36 1.87 0.06 24.93 0.00 67.91

4.28 0.04Semantic category

Natural objects 10 388 0.45 0.29 0.62 5.39 0.00 19.67 0.02 54.25Artifacts objects 9 398 0.35 0.16 0.55 3.54 0.00 16.63 0.03 51.89

0.59 0.44Object recognition task

Naming task 22 851 0.36 0.24 0.48 5.81 0.00 46.31 0.00 54.65Semantic classification task 5 267 0.23 0.00 0.46 1.92 0.06 11.06 0.03 63.84Name-object verification task 7 402 0.11 −0.04 0.26 1.43 0.15 12.05 0.06 50.21

6.46 0.04Stimuli type

Line-drawings 13 514 0.35 0.19 0.50 4.31 0.00 26.48 0.01 54.68Photographs 12 448 0.34 0.18 0.50 4.24 0.00 21.85 0.03 49.65

0.001 0.98Surface details

Present surface details 20 995 0.26 0.13 0.38 4.08 0.00 49.06 0.00 61.27Absent surface details 4 91 0.44 0.22 0.66 3.93 0.00 2.35 0.50 0.00

2.06 0.15Snodgrass and Vanderwart set (1980)

Snodgrass and Vanderwart (1980) set 7 340 0.43 0.25 0.60 4.84 0.00 13.23 0.07 47.10Other object sets 9 393 0.24 0.01 0.27 2.13 0.00 11.18 0.19 28.42

6.83 0.01

Effect sizes (d), 95% confidence intervals, Z-value and significance level (p) for eachmeta-analysis, number of the independent effect sizes (studies or sub-studies) that contributed toeach meta-analysis (Ne), the total number of subjects these effect sizes were based upon (Ns), within-group homogeneity of variance (Qwithin) and significance level (p), percentageof the variation across studies that is due to heterogeneity (I2), between-group homogeneity of variance (Qbetween) and significance level (p).

7I. Bramão et al. / Acta Psychologica xxx (2011) xxx–xxx

3.6. Surface details

The estimated color effect from studies that used colored stimulitogether with other surface details was compared with studies thatused stimuli without surface details. No difference was found(Qbetween=2.06, p=0.15), although both types of studies showedsignificant color effects (surface details present d=0.26, pb0.01;surface details absent d=0.44, pb0.01).

3.7. Snodgrass and Vanderwart set (1980)

To evaluate the moderator effect of the Snodgrass and Vanderwartset (1980), we estimated an effect size of color information fromstudies that used these stimuli and compared with studies that usedother object sets. The color advantage is greater for the Snodgrass andVanderwart set compared to other object sets (Qbetween=6.83,p=0.01). Although both types of studies were associated withsignificant effect sizes, a larger effect for those studies using theSnodgrass and Vanderwart set (d=0.43, pb0.01) was observedcompared to studies that used other object sets (d=0.14, pb0.01).

4. Discussion

In this meta-analytic review of 35 independent experiments, weintended to clarify the role of color information in object recognition.The overall meta-analysis unambiguously revealed that, in contrast tooccasional declarations to the contrary (Biederman & Ju, 1988), colorinformation improves object recognition. This result suggests thatobject recognition theories should consider the role of colorinformation and elaborate its role in object recognition. Moreover, ifwe consider the evolution of the human species, then color visionmost likely developed for specialized uses, including detecting ripefruit amongst foliage (Gegenfurtner, 2003; Surridge, Osorio, &Mundy,2003). Taking such considerations together with the fact that colorplays a prominent part in our subjective experience of the visual

Please cite this article as: Bramão, I., et al., The role of color information(2011), doi:10.1016/j.actpsy.2011.06.010

world, it would make sense to include color processing, and itsconsequences, as an integral part of models of object recognition(Tanaka et al., 2001).

The results of meta-analysis show that the contribution of color inobject recognition depends on object properties and task conditions.In particular, color information facilitates the recognition of colordiagnostic objects and non-color diagnostic objects, albeit less so. Thisresult is not consistent with the color diagnosticity hypothesis ofTanaka and Presnell (1999; see also Nagai & Yokosawa, 2003), thatproposed that color only improves the recognition of high colordiagnostic objects. In the literature review, we identified evidenceshowing a color advantage both for color and non-color diagnosticobjects (Biederman & Ju, 1988; Bramão, Faísca, Petersson, & Reis,2010; Rossion & Pourtois, 2004; Uttl et al., 2006; Wurm et al., 1993).Still, the studies that report a color advantage for non-color diagnosticobjects showed a greater effect for color diagnostic objects (see, forexample, Rossion & Pourtois, 2004). This observation is consistentwith our meta-analytic result: a stronger color effect for colordiagnostic than for non-color diagnostic objects. Moreover, in arecent study (Bramão et al., 2011), we observed that color effects arerestricted to early stages of the visual processing for the non-colordiagnostic objects. Therefore, the failure to find strong color effects fornon-color diagnostic objects in some studies could be related to thenature of the recognition tasks. For instance, it is possible that thecolor effect in non-color diagnostic object recognition is evident onlywhen the recognition task is perceptually sufficiently demanding. It isalso important to note that the Q statistic indicates that the effect sizeestimate for these two groups of studies are not homogenousreflecting high variability between studies (color diagnostic objectsQwithin=24.12, p=0.01; non-color diagnostic objects Qwithin=24.93,pb0.01). However, the limited number of studies did not allow us tofurther investigate what other variables could account for thisheterogeneity.

Several methods have been used to define the color diagnosticityof a particular object. In general, research on the contribution of color

on object recognition: A review and meta-analysis, Acta Psychologica

8 I. Bramão et al. / Acta Psychologica xxx (2011) xxx–xxx

to object recognition would benefit from a standardization of themethod to determine color diagnosticity. Tanaka and Presnell (1999)assessed color diagnosticity using two criteria: feature listing andtypicality judgments. In the feature-listing task, the subjects listedperceptual features associated with an object. In the typicality task,the subjects were asked to indicate the color that was most typical ofan object. An object was rated as high in color diagnosticity only if aspecific color was consistently mentioned first in the feature list andwas rated as the typical color. This approach might be too strict andmight prevent objects that are moderately or even strongly associatedwith a specific color from being classified as color diagnostic objects.Moreover, the feature listing task assesses which properties are moretypical and distinct for an object and not if this object is highlyassociated with a particular color or not. Rossion and Pourtois (2004)used a much more straightforward approach to evaluate colordiagnosticity. They presented each colored object to a group ofsubjects and asked them to rate the object according to the followinginstruction: “give a score between 1 (the color of the presented objectis not diagnostic at all, i.e., the object could be in any other colorequally well) and 5 (the color is highly diagnostic of the object, i.e., theobject appears only with that color in real life).” We suggest thatmethods, similar to the one of Rossion and Pourtois (2004), are morerelevant and carries greater face validity in assessing whether aparticular object is associated with a specific color, or not. In reality,the association between an object and a color (or range of colors) is amatter of degree, determined by several factors including statisticalfrequencies. Moreover, we also think that it is important in thiscontext is to ensure that the subjects classify the object concept asbeing associated with a color (or not) and not to classify the objectimage itself. This avoids the possibility that results are dependent on aparticular set of images, making the results of these types of studiesmore comparable. Another important consideration is thatmost of thestudies that have explored the role of color diagnosticity in objectrecognition classify the objects as being either color diagnostic or non-color diagnostic (see, for an exception, Rossion & Pourtois, 2004). Webelieve that color diagnosticity is better described as a continuousvariable, with high color diagnostic objects lying at one end of thecontinuum, the non-color diagnostic objects at the other, and theobjects with moderate color associations localized somewhere inbetween.

The meta-analytic results also showed a similar color advantageeffect for studies that used natural objects and artifacts. This result isconsistent with several previous studies (Bramão, Faísca, Petersson, &Reis, 2010; Rossion & Pourtois, 2004; Uttl et al., 2006). However, it isimportant to note that the Q statistic indicates that the effect sizeestimates are non-homogenous, reflecting high variability (naturalobjects Qwithin=19.67, p=0.02; artifacts Qwithin=16.63, p=0.03).Going forward, it is important to determine which other variables arecontributing to this variability. For example, to cross the semanticcategory variable with the color diagnostic variable. However, thelimited number of studies available prevented us from pursuing thisanalysis.

The meta-analysis also revealed that color contributes to objectrecognition mainly in studies that used naming tasks but less so instudies that used semantic classification tasks. Surprisingly, thestudies that used verification tasks did not show a significant coloreffect. These tasks also require semantic and name activation similarto the naming task. In the literature review, we found a robust body ofevidence showing that color information improves object recognitionin object and scene verification tasks; however, the meta-analysis didnot replicate this finding. One possible explanation for the discrep-ancy might be related to the color diagnosticity factor. On closeexamination, out of the seven studies that employed a verificationtask, only two experiments used only color diagnostic objects (Joseph,1997; Therriault et al., 2009). Two other used both color and non-color diagnostic objects (Biederman & Ju, 1988), and in the other three

Please cite this article as: Bramão, I., et al., The role of color information(2011), doi:10.1016/j.actpsy.2011.06.010

experiments we only considered the non-color diagnostic objectsresults (to guarantee that effect size values came from independentstudies) (Bramão, Faísca, Petersson, & Reis, 2010; Tanaka & Presnell,1999). It is also relevant to note that the effect sizes in studies thatused naming and semantic classification tasks are non-homogeneous,which suggests that other sources are causing the between-studyvariability (naming task Qwithin=46.31, pb0.01; semantic classifica-tion task Qwithin=11.06, p=0.03).

To fully understand the level of processing at which colorinformation improves object recognition, it is important to understandwhich object recognition tasks benefit (or not) from color information.In the literature review, we also found evidence that color informationimproves object naming to a greater degree than semantic classifica-tions (Biederman & Ju, 1988; Davidoff & Ostergaard, 1988). In fact,Davidoff (1991) proposed a model of object recognition where colorinformation only plays a role in the later stages of the visual processing,after the structural description of the object has been computed. In thismodel, the input is analyzed by a boundary feature contour system, theresulting representation of information is stored temporary, andactivates a stored structural description which is spatially defined byshape and size information, but not by color. The identification of thestructural description then activates associated stored knowledge forthat particular object. Davidoff (1991) considered two basic forms ofstored knowledge: the hasa knowledge concerning sensory informa-tion and the isa knowledge concerning information about the functionof the object. Object color, according to this model, is specifically partof the associated hasa properties. So, the color effects in objectrecognition are considered to take place after the initial visualrepresentation. In concordance with the prediction that colorinformation is not available in the structural descriptions, Davidoffand Ostergaard (1988) found that the introduction of color did notimprove performance in a size comparison task.

The absence of color effects at the level of stored structuraldescriptions was first disputed by Price and Humphreys (1989), whoargued that color is required at the structural description stage todisambiguate objects from categories that are structurally similar. Priceand Humphreys (1989) proposed that there are separated representa-tions for color and shape, and that these are intimately interconnected.Appropriately colored objects activate color representations that inturn activate associated shape representations (Humphreys, Goodale,Jakobson, & Servos, 1994; Price & Humphreys, 1989). Consistent withthis, it has been reported that when the correlation between color andshape is high, as it is in color diagnostic objects, thepresenceof color aidsrecognition to a greater degree than when the correlation betweenshape and color is low (see, for example, Rossion & Pourtois, 2004). In arecent study, we showed that the role of color in object recognition isdependent on the color diagnosticity status of objects (Bramão et al.,2011). For the recognition of the color diagnostic objects, colorinformation was especially important at the level of semantic represen-tation, whereas for non-color diagnostic objects, color informationimproved object recognition at early stages of the visual processingsuggesting that color improves object recognition at early stages of thevisual processing for all objects. However, since non-color diagnosticobjects are not strongly associated with a color, no further coloradvantage is expected at the higher processing levels.

Moreover, based on the literature review, it appears that colorinformation contribute inmore ways to object recognition than just toprovide a link between the object shape and the object namerepresentation, or to facilitate semantic access for color diagnosticobjects. The studies included in the meta-analysis mostly investigatednaming, object verification and semantic classification. This ruled outthe possibility to test whether color has a role at the early processingstages of object recognition. However, there is evidence suggestingthat this is the case (Davidoff et al., 1997; Gegenfurtner & Rieger,2000; Wurm et al., 1993). For example, Gegenfurtner and Riegertested the role of color vision in the recognition of briefly presented

on object recognition: A review and meta-analysis, Acta Psychologica

9I. Bramão et al. / Acta Psychologica xxx (2011) xxx–xxx

images of natural scenes using a delayed match-to-sample task. Theresults showed a clear, and rapid, effect of color on the recognition ofnatural images in at least two ways: at an early stage, where colorcontributes with an additional cue for image segmentation; and at alater stage, where color served as a cue for memory retrieval(Gegenfurtner & Rieger, 2000). Further research is needed to clarifythe role of color and at which processing stages color informationmodulates object recognition.

Additional moderator variables related to the effects of stimuluscharacteristics (stimulus type and surface details) were explored inthe meta-analysis. We found significant color effects independent ofthese characteristics. Thus, it seems that color plays a similar role in therecognition of types (line-drawings) and tokens (photographs).Moreover, it appears that color alone, without additional surfacedetails, unambiguously improves object recognition, which is consis-tent with the findings of the literature review (Bramão et al., 2011;Joseph, 1997; Moore & Price, 1999; Price & Humphreys, 1989; Reiset al., 2006; Vernon & Lloyd-Jones, 2003). The homogeneity measures(Q and I2) indicated that the effect size for these groups of studieswerenon-homogenous, suggesting that other variables contribute to thisvariability. However, the limited number of studies does not allow usto further explore other potential sources of variability. Interestingly,we found a superior color advantage for studies that used theSnodgrass and Vanderwart (1980) set or its colored version (fromRossion&Pourtois, 2004). This advantagemight be due to the set itself,for instance theway the objects are drawn, painted, or displayed. Lawsand colleagues (Laws, Adlington, Gale, Moreno-Martínez, & Sartori,2007) point out that normal subjects often perform near ceiling on theSnodgrass and Vanderwart set in standard paradigms, which mightlead to a lack of sensitivity between conditions or groups in objectnaming studies. To solve this problem, a new set of less familiar objectswere constructed (Adlington, Laws, & Gale, 2009a).

5. Conclusion

In summary, both the literature review and the meta-analysissuggested that color information contributes to object recognition, inparticular for objects strongly associated with a color, but also forobjects not so strongly associated with a particular color. Colorinformation also improves the recognition of natural objects andartifacts, to a similar degree, as well as the recognition of tokens(photographs) and types (line-drawings). Moreover the color advan-tage effect seems to be independent of other surface details, includingshadows and texture. Finally, in almost all the subsets of the includedstudies, the effect size estimates are heterogeneous, with theexception of studies that used images without surface details andstudies not using the Snodgrass and Vanderwart (1980) set. It is animportant scientific task to explore the underlying sources ofheterogeneity observed in future research of color processing.

Acknowledgment

This work was supported by Max Planck Institute for Psycholin-guistics, Donders Institute for Brain, Cognition and Behaviour,Stockholm Brain Institute, Fundação para a Ciência e Tecnologia(FCT/POCTI/46955/PSI/2002; PTDC/PSI-PCO/110734/2009; IBB/CBME,LA, FEDER/POCI 2010), Vetenskapsrådet, Hedlunds Stiftelse, StockholmCounty Council (ALF, FoUU), and a PhD fellowship awarded to InêsBramão (FCT/SFRH/BD/27381/2006).

References

Adlington, R., Laws, K., & Gale, T. (2009). The Hatfield Image Test (HIT): A new picturetest and norms for experimental and clinical use. Journal of Clinical and ExperimentalNeuropsychology, 31, 731–753.

Please cite this article as: Bramão, I., et al., The role of color information(2011), doi:10.1016/j.actpsy.2011.06.010

Adlington, R., Laws, K., & Gale, T. (2009). Visual processing in Alzheimer's disease: Surfacedetail and colour fail to aid object identification. Neuropsychologia, 47, 2574–2583.

Alario, F. X., & Ferrand, L. (1999). A set of 400 pictures standardized for French: Normsfor name agreement, image agreement, familiarity, visual complexity, imagevariability, and age of acquisition. Behavior Research Methods, Instruments, &Computers, 31, 531–552.

Biederman, I. (1987). Recognition-by-components: A theory of human imageunderstanding. Psychological Review, 94, 115–147.

Biederman, I., & Ju, G. (1988). Surface versus edge-based determinants of visualrecognition. Cognitive Psychology, 20, 38–64.

Bonin, P., Peereman, R., Malardier, N., Méot, A., & Chalard, M. (2003). A new set of 299pictures for psycholinguistic studies: French norms for name agreement, imageagreement, conceptual familiarity, visual complexity, image variability, age ofacquisition, and naming latencies. Behavior Research Methods, Instruments, &Computers, 35, 158–167.

Boucart,M., Despretz, P., Hladiuk, K., &Desmettre, T. (2008). Does context or color improveobject recognition iin patients with low vision? Visual Neuroscience, 25, 685–691.

Bramão, I., Faísca, L., Forkstam, C., Reis, A., & Petersson, K. M. (2010). Cortical brainregions associated with color processing: An FMRI study. The Open NeuroimagingJournal, 4, 164–173.

Bramão, I., Faísca, L., Petersson, K. M., & Reis, A. (2010). The influence of surface colorinformation and color knowledge information in object recognition. The AmericanJournal of Psychology, 123, 459–468.

Bramão, I., Inácio, F., Faísca, L., Reis, A., & Petersson, K. M. (2011). The influence of colorinformation on the recognition of color diagnostic and non color diagnostic objects.The Journal of General Psychology, 138, 1–17.

Brodie, E., Wallace, A., & Sharrat, B. (1991). Effect of surface characteristics and style ofproduction on naming and verification of pictorial stimuli. The American Journal ofPsychology, 105, 517–545.

Chao, L., & Martin, A. (1999). Cortical regions associated with perceiving, naming, andknowing about colors. Journal of Cognitive Neuroscience, 11, 25–35.

Cuetos, F., Ellis, A. W., & Alvarez, B. (1999). Naming times for the Snodgrass andVanderwart pictures in Spanish. Behavior Research Methods, Instruments, &Computers, 31, 650–658.

Davidoff, J. (1991). Cognition Through Color. Cambridge: MIT Press.Davidoff, J., & Ostergaard, A. (1988). The role of colour in categorial judgement.

Quarterly Journal of Experimental Psychology, 40, 533–544.Davidoff, J., Walsh, V., & Wagemans, J. (1997). Higher-level cortical processing of color.

Acta Psychologica, 97, 1–6.Dell'Acqua, R., Lotto, L., & Job, R. (2000). Naming times and standardized norms for the

Italian PD/DPSS set of 266 pictures: Direct comparisons with American, English,French, and Spanish published databases. Behavior Research Methods, Instruments,& Computers, 32, 588–615.

Faísca, L., Mendonça, A., Bramão, I., Ingvar, M., Petersson, K. M., & Reis, A. (2004).Demographic and stimulus attribute effects on object naming. Paper presented atthe First Portuguese Forum of Experimental Psychology, Universidade do Minho, Braga.

Gale, T., Laws, K., & Foley, K. (2006). Crowded and sparse domains in object recognition:Consequences for categorization and naming. Brain and Cognition, 60, 139–145.

Gegenfurtner, K. (2003). Cortical mechanisms of colour vision. Nature ReviewsNeuroscience, 4, 563–572.

Gegenfurtner, K., & Rieger, J. (2000). Sensory and cognitive contributions of color to therecognition of natural scenes. Current Biology, 10, 805–808.

Hocking, J., & Price, C. (2008). The influence of colour and sound on neuronal activationduring visual object naming. Brain Research, 1241, 92–102.

Humphreys, G.W., Goodale, M. A., Jakobson, L. S., & Servos, P. (1994). The role of surfaceinformation in object recognition: Studies of a visual form agnosic and normalsubjects. Perception, 23, 1457–1481.

Humphreys, G. W., Price, C., & Riddoch, M. J. (1999). From objects to names: A cognitiveneuroscience approach. Psychological Research, 62, 118–130.

Joseph, J. (1997). Color processing in object verification. Acta Psychologica, 97, 95–127.Joseph, J., & Proffitt, D. (1996). Semantic versus perceptual influences of color in object

recognition. Journal of Experimental Psychology. Learning, Memory, and Cognition, 22,407–429.

Laws, K., Adlington, R., Gale, T., Moreno-Martínez, F., & Sartori, G. (2007). A meta-analyticreview of category naming in Alzheimer's disease. Neuropsychologia, 45, 2674–2682.

Laws, K., & Hunter, M. Z. (2006). The impact of colour, spatial resolution, andpresentation speed on category naming. Brain and Cognition, 62, 89–97.

Lloyd-Jones, T., & Nakabayashi, K. (2009). Independent effects of colour on objectidentification and memory. The Quarterly Journal of Experimental Psychology, 62,310–322.

Mapelli, D., & Behrmann, M. (1997). The role of color in object recognition: Evidencefrom visual agnosia. Neurocase, 3, 237–247.

Marr, D., & Nishihara, H. (1978). Representation and recognition of the spatialorganization of three-dimensional shapes. Proceedings of the Royal Society ofLondon, Series B, 200, 269–294.

Moore, C., & Price, C. (1999). A functional neuroimaging study of the variables thatgenerate category-specific object processing differences. Brain, 122, 943–962.

Nagai, J., & Yokosawa, K. (2003). What regulates the surface color effect in objectrecognition: Color diagnosticity or category? Technical Report on Attention andCognition, 28, 1–4.

Naor-Raz, G., & Tarr, M. J. (2003). Is color an intrinsic property of object representation?Perception, 32, 667–680.

Oliva, A., & Schyns, P. G. (2000). Diagnostic colors mediate scene recognition. CognitivePsychology, 41, 176–210.

Ostergaard, A., & Davidoff, J. (1985). Some effects of color on naming and recognition ofobjects. Journal of Experimental Psychology. Learning,Memory, andCognition,11, 579–587.

on object recognition: A review and meta-analysis, Acta Psychologica

10 I. Bramão et al. / Acta Psychologica xxx (2011) xxx–xxx

Price, C., & Humphreys, G.W. (1989). The effects of surface detail on object categorizationand naming. The Quarterly Journal of Experimental Psychology, 41, 797–827.

Reis, A., Faísca, L., Ingvar, M., & Petersson, K. M. (2006). Color makes a difference: Two-dimensional object naming in literate and illiterate subjects. Brain and Cognition,60, 49–54.

Rosenthal, R. (1991). Meta-Analytic Procedures for Social Research. London: SagePublications.

Rossion, B., & Pourtois, G. (2004). Revisiting Snodgrass and Vanderwart's object pictorialset: The role of surface detail in basic-level object recognition. Perception, 33, 217–236.

Ryan, C. S., Hemmes, N. S., & Brown, B. L. (2003). The effect of chromaticity varies withobject identification response: speeded naming versus recognition. PsychologicalRecord, 53, 467–486.

Sanfeliu, M. C., & Fernandez, A. (1996). A set of 254 Snodgrass–Vanderwart picturesstandardized for Spanish: Norms for name agreement, image agreement,familiarity, and visual complexity. Behavior Research Methods, Instruments, &Computers, 28, 537–555.

Snodgrass, J. G., & Vanderwart, M. (1980). A standardized set of 260 pictures: Norms forname agreement, image agreement, familiarity, and visual complexity. Journal ofExperimental Psychology. Learning, Memory, and Cognition, 6, 174–215.

Please cite this article as: Bramão, I., et al., The role of color information(2011), doi:10.1016/j.actpsy.2011.06.010

Surridge, A., Osorio, D., & Mundy, N. (2003). Evolution and selection of trichromaticvision in primates. Trends in Ecology & Evolution, 18, 198–205.

Tanaka, J., & Presnell, L. (1999). Color diagnosticity in object recognition. Perception &Psychophysics, 61, 1140–1153.

Tanaka, J., Weiskopf, D., & Williams, P. (2001). The role of color in high-level vision.Trends in Cognitive Sciences, 5, 211–215.

Therriault, D. J., Yaxley, R. H., & Zwaan, R. A. (2009). The role of color diagnosticity inobject recognition and representation. Cognitive Processing, 10, 335–342.

Uttl, B., Graf, P., & Santacruz, P. (2006). Object color effects identification and repetitionpriming. Scandinavian Journal of Psychology, 47, 313–325.

Ventura, P. (2003). Normas para figuras do corpus de Snodgrass e Vanderwart (1980).Laboratório de Psicologia, 1, 5–19.

Vernon, D., & Lloyd-Jones, T. (2003). The role of the colour implicit and explicit memoryperformance. The Quarterly Journal of Experimental Psychology, 56A, 779–802.

Wurm, L. H., Legge, G. E., Isenberg, L. M., & Luebker, A. (1993). Color improves objectrecognition in normal and low vision. Journal of Experimental Psychology. HumanPerception and Performance, 19, 899–911.

on object recognition: A review and meta-analysis, Acta Psychologica