{"@context":{"@language":"en","Affiliation":"http:\/\/vivoweb.org\/ontology\/core#departmentOrSchool","AggregatedSourceRepository":"http:\/\/www.europeana.eu\/schemas\/edm\/dataProvider","Campus":"https:\/\/open.library.ubc.ca\/terms#degreeCampus","Creator":"http:\/\/purl.org\/dc\/terms\/creator","DateAvailable":"http:\/\/purl.org\/dc\/terms\/issued","DateIssued":"http:\/\/purl.org\/dc\/terms\/issued","Degree":"http:\/\/vivoweb.org\/ontology\/core#relatedDegree","DegreeGrantor":"https:\/\/open.library.ubc.ca\/terms#degreeGrantor","Description":"http:\/\/purl.org\/dc\/terms\/description","DigitalResourceOriginalRecord":"http:\/\/www.europeana.eu\/schemas\/edm\/aggregatedCHO","FullText":"http:\/\/www.w3.org\/2009\/08\/skos-reference\/skos.html#note","Genre":"http:\/\/www.europeana.eu\/schemas\/edm\/hasType","IsShownAt":"http:\/\/www.europeana.eu\/schemas\/edm\/isShownAt","Language":"http:\/\/purl.org\/dc\/terms\/language","Program":"https:\/\/open.library.ubc.ca\/terms#degreeDiscipline","Provider":"http:\/\/www.europeana.eu\/schemas\/edm\/provider","Publisher":"http:\/\/purl.org\/dc\/terms\/publisher","Rights":"http:\/\/purl.org\/dc\/terms\/rights","ScholarlyLevel":"https:\/\/open.library.ubc.ca\/terms#scholarLevel","Title":"http:\/\/purl.org\/dc\/terms\/title","Type":"http:\/\/purl.org\/dc\/terms\/type","URI":"https:\/\/open.library.ubc.ca\/terms#identifierURI","SortDate":"http:\/\/purl.org\/dc\/terms\/date"},"Affiliation":[{"@value":"Arts, Faculty of","@language":"en"},{"@value":"Psychology, Department of","@language":"en"}],"AggregatedSourceRepository":[{"@value":"DSpace","@language":"en"}],"Campus":[{"@value":"UBCV","@language":"en"}],"Creator":[{"@value":"Wig, Patrick Shawn","@language":"en"}],"DateAvailable":[{"@value":"2011-02-11T23:54:02Z","@language":"en"}],"DateIssued":[{"@value":"1991","@language":"en"}],"Degree":[{"@value":"Master of Arts - MA","@language":"en"}],"DegreeGrantor":[{"@value":"University of British Columbia","@language":"en"}],"Description":[{"@value":"Search for combinations of orientation, shape, color, direction of motion, and binocular disparity should be slower than search for any one of them in isolation. Recent results showing fast search for certain combinations of these features havef prompted theorists to suggest that features like orientation may be represented separately depending on the color, texture, or other medium that distinguishes them from the background (Cavanagh, Arguin, & Treisman, 1990; Treisman, 1988). Enns and Wig (1989) provide evidence that edge orientation is represented independently from the medium in which it is carried, at least for texture and luminance. This thesis examined the effects of multiple colors on search for oriented edges. Experiment 1 compared edges defined by different equiluminant colors (i.e., colors of equal apparent brightness), and Experiment 2 compared color and luminance defined edges. Subjects were able to search for a vertical edge defined by either color in parallel. They could also search for a vertical edge of randomly alternating color among mixed color horizontal distractors without interference. However, search rates became serial if the distractors included vertical edges of another color. This pattern of results was duplicated in Experiment 2. The presence of this Stroop interference is\r\nstrong evidence for color and luminance independent representations for orientation in visual search.","@language":"en"}],"DigitalResourceOriginalRecord":[{"@value":"https:\/\/circle.library.ubc.ca\/rest\/handle\/2429\/31222?expand=metadata","@language":"en"}],"FullText":[{"@value":"SEARCH FOR ORIENTATION IN MULTIPLE MEDIA: WHICH COMES FIRST, THE COLOR OR THE EDGE by PATRICK SHAWN WIG B. A., The University of Saskatchewan, 1988 A THESIS SUBMITTED IN PARTIAL FULFILLEMNT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ARTS in THE FACULTY OF GRADUATE STUDIES (Department of Psychology) We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA August 1991 \u00a9 Patrick Shawn Wig, 1991 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of Psychology The University of British Columbia Vancouver, Canada Date 7 OCT 91 . DE-6 (2\/88) Abstract Search for combinations of orientation, shape, color, direction of motion, and binocular disparity should be slower than search for any one of them in isolation. Recent results showing fast search for certain combinations of these features havef prompted theorists to suggest that features like orientation may be represented separately depending on the color, texture, or other medium that distinguishes them from the background (Cavanagh, Arguin, & Treisman, 1990; Treisman, 1988). Enns and Wig (1989) provide evidence that edge orientation is represented independently from the medium in which it is carried, at least for texture and luminance. This thesis examined the effects of multiple colors on search for oriented edges. Experiment 1 compared edges defined by different equiluminant colors (i.e., colors of equal apparent brightness), and Experiment 2 compared color and luminance defined edges. Subjects were able to search for a vertical edge defined by either color in parallel. They could also search for a vertical edge of randomly alternating color among mixed color horizontal distractors without interference. However, search rates became serial if the distractors included vertical edges of another color. This pattern of results was duplicated in Experiment 2. The presence of this Stroop interference is i i i strong evidence for color and luminance independent representations for orientation in visual search. i v Table of Contents Abstract ...i i Table of Contents iv List of Tables vi List of Figures vi i Acknowledgement .' v i i i Search for orientation in multiple media Which comes first, the color or the edge? 1 Theories of early vision 4 Medium and message in early vision 9 Research treatment of the medium\/message distinction 12 Interference tasks Finding orientations in multiple colors 1 5 Stroop interference 1 5 Garner Orthogonal Set interference 1 6 Previous related research 18 Experiment 1 23 Method 23 Design 23 Subjects 25 Stimuli 25 Procedure 27 Color equi-brightness determination 28 Results 31 V Stroop task 31 Garner task 3 4 Discussion 37 Experiment 2 3 8 Method 4 0 Results 40 Stroop task 4 0 Garner task 4 5 Discussion, 48 General Discussion 1 49 References 5 3 Figure Captions 58 List of Tables Table 1: Color Brightness Levels and Fusion Ranges as a Function of Screen Masking v i i List of Figures Figure 1. Treisman's feature integration theory 61 Figure 2. Surface media which carry edges 62 Figure 3. Treisman's multiple map theory 63 Figure 4. Stroop interference paradigm 64 Figure 5. Garner interference paradigm 65 Figure 6. Stroop and Garner task conditions 66 Figure 7. Fusion ranges for color choices 67 Figure 8. Experiment 1 Stroop reaction times and errors 68 Figure 9. Experiment 1, Stroop slopes 69 Figure 10. Experiment 1 Garner reaction times and errors 70 Figure 11. Experiment 1 Garner slopes 71 Figure 12. Experiment 2 Stroop reaction times and errors 72 Figure 13. Experiment 2 Stroop slopes 73 Figure 14. Experiment 2 Garner reaction times and errors 74 Figure 15. Experiment 2 Garner slopes 75 Acknowledgement The author is indebted to Dr. J . Enns for inspiration and for invaluable technical, methodological and editorial guidance. Debts are also owed to Dr. R. Lakowski for additional technical and methodological assistance, and to Dr. C. Rankin for structural revisions to this thesis. This research was funded in part by Post Graduate Fellowships from the Natural Sciences and Engineering Council of Canada, and University Graduate Fellowships from the University of British Columbia. 1 Search for orientation in multiple media: Which comes first, the color or the edge? When presented with a visual display containing several stimuli, subjects report that certain differences appear to \"pop out\". These differences have been called preattentive features and are thought to be the basic building blocks of visual perception. Treisman's feature integration theory (Treisman & Gelade, 1980) interprets and predicts the circumstances under which these differences can be detected. For the most part, combinations of features cannot be detected without effortful serial search. There are, however, some combinations which do pop out, causing researchers to question the completeness of the feature integration theory. Treisman (1988) has noted that in cases where combinations of two or more features pop out, one of the features is usually a surface property like color, luminance, texture, and relative motion. The other feature is often one that characterizes the edges and the general shape of objects. These are features like orientation, shape, curvature, and length. Treisman (1988) posits a modified feature integration theory in which attention prevents accidental combinations of features defined by a particular surface difference such as color, and is not needed to prevent combinations across categories like the orientation of an object and the orientations of its surfaces, 2 or between the orientations of two objects defined by different color or texture boundaries. In broader strokes, this modification allows the feature integration model to keep separate collections of shape and orientation features for each surface feature distinction present in a display. This modification would predict the pop out of any combination of two features from different categories, by allowing parallel search within the relevant surface medium without interference from features in other surface media. This theory, however, results in an unnecessary proliferation of feature maps for each surface medium in a complex visual display. A categorization of shape, surface, and surface boundary features regardless of how or where they are represented may be a more frugal means of feature extraction. In terms of visual function, the detection of a particular edge orientation like a horizontal visual cliff, whether defined by color, luminance, texture, binocular disparity, relative motion, or any combination of them, may be much more important than knowing which of these media carry the edge. This more abstract map system provides just the sort of fast orientation extraction required for ecological vision. Theoretically, such an abstract set of maps may dramatically reduce and simplify the feature extraction process, and therefore make search among these maps much simpler and 3 faster than in Treisman's (1988) modified theory. The cost of such a system would, however, be a failure to explain parallel search for feature combinations, at least as long as we continue to propose that each map is discrete and incapable of sharing information preattentively. This thesis will examine the relation between orientation representation in the visual system, and the colors in which this feature is presented. If the same orientation is present in several colors, does the visual system categorize this feature separately for each color, or does it categorize the orientation only once and tag all instances regardless of color? This is an important question because different answers have very different implications for the organization of the visual system. Color-independent representations of features like orientation suggest a much more complex and abstract initial categorization than would the series of color-dependent representations suggested by the current feature integration theory. As background for this question I will first outline currently used general theories of preattentive vision, and recent anomalous data which raise relevant questions about them. I will then define the terms medium and message and outline how their application may clarify these questions. The background for questioning medium and message interactions in visual processing will be presented, and a means of answering 4 this question will be discussed. Finally, two experiments will be presented to answer the specific questions raised about medium and message interactions in the human visual system. Theories of early vision Current visual processing theories postulate two sub systems: preattentive and attentive. Attentive processes are selective and resource limited. At the risk of being inundated with information from the visual field, the attentive system actively selects those aspects which have been chosen or defined to be useful, and devotes time and effort to their processing. There do, however, appear to be processes which are not selective or resource limited, and which appear to be performed without the involvement of attention. These processes appear to be effortlessly performed in a parallel manner across the visual field before attention is actually involved. These latter processes are generally called preattentive. Within preattentive vision research one of the foci has been the identification of visual primitives. The visual search paradigm is one of several tools used to this end. In this task, subjects are presented with displays of various numbers of stimuli either including or not including a predefined target stimulus. Subjects must then decide as quickly as possible whether the target stimulus is present or absent; differences between stimuli which allow their immediate and parallel 5 detection across the visual field, regardless of the number of items present, are believed to be.primitive features. Features found in this way include color, presence of line terminators, direction of motion, closure, orientation, and size (Cavanagh, et al. 1990; Treisman & Gelade, 1980), binocular disparity (Nakayama & Silverman, 1986), and spatial frequency (Beck, Sutter, & Ivry, 1987). All of these characteristics have a compelling ecological validity, and may be readily recognized as useful for distinguishing objects, forms, textures, and distance. Treisman's feature integration theory (Cavanagh, et al. 1990; Treisman; 1986; Treisman, 1988; Treisman & Gormican, 1988) proposes that in addition to abstracting these features from the visual field in parallel, the visual system categorizes them. This categorization takes place by the coding of specific features within groups of maps called modules (see Figure 1). Each module codes information of a particular sort: colors, orientations, sizes, and so forth. Within each of these modules, individual maps code features differing from each other along that particular dimension. Where these dimensions are continuous, the maps may themselves be continuous, with widely disparate entries being functionally discrete. Dimensions of a discrete nature, shape or sign of curvature for example, may be coded as several individual and discrete maps. These maps are proposed to be location insensitive, so activation in any one of 6 them may arise from anywhere in the visual field (within the limits of visual acuity). This proposed system structure predicts its ability to detect the presence of any single sufficiently distinguishable feature in parallel across the visual field by monitoring the requisite feature map. Insert Figure 1 about here According to this theory, the visual system acquires information about combinations of features by bringing the spotlight of attention to a particular point on a master map of locations. This master map contains information only about the presence or absence of feature boundaries, and no information about which particular features are involved. When the spotlight is brought to a particular point, all features active there are automatically retrieved from their separate map locations and can be combined. It is this attentional feature integration which gives the theory its name. This aspect of the theory also describes its functional limit. Two or more features can not be discriminated in combination without attention. Any distinction in the visual field which requires such a combination or conjunction of features cannot be performed in parallel, but must be made by serial search through the relevant items by the \"spotlight of attention\" (Cavanagh, et al. 1990; Treisman, 1986; Treisman, 1988; Treisman & Gelade, 1980; Treisman & Gormican, 1988), before the task can be performed. Based on this theoretical distinction, the presence of a visual feature should be detected in parallel across the visual field, while combinations or conjunctions of features should require a serial search of individual items or small groups of items in order to detect their presence. Before describing the current research question, and the means by which it was tested, several definitions are required. Since this thesis questions the representation of features in the visual system, a definition of the term \"feature\" is in order. Visual features are theoretical constructs believed to be the most basic, primary element codes in the human visual system. They are the first units of information abstracted from a scene, and the first categorized. All further visual information is derived from and discriminated by combinations of these feature elements. These feature elements are also believed to be abstracted from large areas of the visual field, simultaneously categorized into separate coexisting maps, and referenced to a single location map for the entire visual field. Given these theoretical properties, a feature can be operationally defined as any difference that the visual system can detect in parallel across the visual field. 8 Within the paradigm of visual search, the terms serial search and parallel search are frequently used. Serial search has been defined by the following characteristics: an increase in response latency of greater than ten milliseconds per item added to the display (ms\/item, a measure of response time slope or RT slope), and a ratio of 2:1 between target absent and target present slopes. Parallel search has conversely been defined by RT slopes of less than 10 ms\/item, a 1:1 ratio of target absent to target present slopes, or both. Recent results raise specific problems for the Feature Integration Theory's depiction of feature maps and the ways the information they contain can be used. Wolfe, Cave, and Franzel (1989) found parallel search rates for conjunctions of color and form (red Os against a background of green Os and red Xs), and of color and orientation (green horizontal lines against red horizontal lines and green vertical lines). In these experiments, saturated reds and greens were used against a black background. When colors were less saturated (Wolfe et al., 1989, Experiment 7), the search slopes increased to ambiguous levels (10.6 ms\/item for target present trials, and approximately a 2:1 absent slope to present slope ratio) but these results were more nearly parallel than similar studies by Treisman and Gelade (1980). Searches for triple conjunctions (i.e., targets defined by the unique combination of three features, where distractors are 9 typically different combinations of these same features) conducted by Wolfe et al., (1989) showed the most striking deviation from Feature Integration Theory predictions. They documented faster conjunction searches for targets differing from distractors by two dimensions in a triple conjunction task, than for targets differing by only one dimension in a double conjunction task. Instead of becoming more difficult and therefore more serial, these searches were easier than simple two feature conjunctions. Triple conjunction searches involving stimuli that differed from the target in only one of the three dimensions were found to be roughly as fast as simple two feature conjunction searches. Rather than supporting separate maps which can not be combined without using a specific local attention spotlight, information from several maps appears available at the same time to allow either parallel search for certain combinations, or fast serial self terminating search through one of the dimensions guided by information available from other relevant maps. Medium and message in early vision The focus of this thesis is on the relationship between various primitive features in early vision. For example, a number of means can be used to define an oriented edge, including luminance, texture, and color contrasts, binocular disparity, and relative motion (see Figure 2). Most researchers have treated these defining surface characteristics as additional features of an object, and not as a dimension independent of feature composition. Insert Figure 2 about here Other than very recent research (Cavanagh, et al. 1990; Enns & Wig, 1989; Treisman & Gormican, 1988; Treisman, 1990) few have asked whether the way in which features are represented (their media) is independent of the feature itself, or whether it is treated by the visual system as part of the feature. To accomplish this end, the terms medium and message must be defined. A useful distinction can be made between the physical world of three-dimensional objects, and the projection of light from that world onto the retina. This external physical world of edges, shapes, contours, and distance will be called the scene. The projection of this scene into the visual system will be referred to as the image. The scene is physically present and can be manipulated or changed, and exists independent of perception. The visual system's representation of the scene is limited by that system's acuity and quality of resolution. Given this distinction, the terms medium and message used throughout the rest of this paper can be made clear. In terms of the visual system, the message is any difference or 11 characteristic present in the scene, which the visual system can resolve and discriminate from others like it. For example, the orientation of a perceivable object edge or boundary can be defined as a message. Other characteristics of objects in the scene, such as their particular color, surface texture, shape, relative motion, or any combination of them, can also be messages. A message is therefore an intrinsic property of the scene. An essential part of this definition of message is that it can be perceived by the visual system. By perception, I mean that there must be some difference or discontinuity in the scene which is reflected in the image, and which can be registered by the visual system. The discontinuity in the image that carries the message is the medium. For example, the oriented edge in the previous example can be represented by any of several image media. A discontinuity of color, spatial frequency, texture, luminance, binocular disparity, apparent or relative motion, reflectance, or any combination of these may be an adequate medium to represent a particular oriented edge. Alternatively, many scene properties such as infra-red luminance and temperature gradients for example, exist with the potential to convey scene information but their presence or absence is imperceptible to the unaided observer, and thus does not fit the definition of message. In a similar respect, any time the medium discontinuity becomes sufficiently weak or noisy to make its distinction impossible, the message is said to be lost. Research treatment of the medium\/message distinction We come now to the question of this thesis. Given that a specific orientation may be presented to the visual system by several colors concurrently, are these instances coded separately or together? Several different backgrounds exist for this question in previous literature. Each deals with the potential distinction of message and media in a slightly different manner. Certain researchers have previously ignored or failed to notice the medium\/message distinction (McLeod, Driver, & Crisp, 1988; Nakayama & Silverman, 1986). Nakayama and Silverman (1986) argued that parallel search rates for conjunctions of binocular disparity (the message) and color (the medium) implied that the visual system used depth to segregate the visual field, and that other features lacked this power. In their studies, relative motion, an alternate medium, did not allow such effortless segregation. McLeod, et al. (1988) showed that with the proper manipulation, combinations of form (the message) and motion (the medium) could be segregated as well as when binocular disparity was used as a medium. These authors have, however, treated potential media (e.g., relative motion, binocular disparity) as separate, independent messages conjoined with others. Where feature conjunctions could be searched for in parallel, such conjunctions had a special status in feature maps, and as such provided arguments against simple multiple map approaches. Conversely, serial results from such conjunction search tasks supported multiple map theories by proving the necessity of attention to serially combine or conjoin features before specific target searches could be made. Both Nakayama and Silverman (1986) and Macleod, et. al. (1988) treated media as additional features of an object, and interpreted any interactions between these media and other features present as interactions between distally represented features in separate maps. Recent research has raised the possibility of separate message maps for each medium present in the representation (Treisman, 1988; see Figure 3a). This research treats medium and message separately, but still treats them like features. Medium becomes another level of feature mapping, and is placed at the lowest or first level in the processing stream. It therefore multiplies all messages by the number of media present. Note in Figure 3a that only the orientation maps have been shown. Current results are ambiguous about where this \"media map\" level belongs (Cavanagh, et al. 1990). Insert Figure 3 about here This distinction, however, does not address the interaction of media with the features represented. These combinations of message and medium have still been treated as conjunctions of features, and as such raise nearly as many difficulties as this new interpretation solves. For example, this interpretation not only fails to explain the apparent information sharing between maps documented by Wolfe, et al. (1989), but also results in a proliferation of redundant \"maps\". In this thesis, I address combinations of media (colors and combinations of color and luminance) and messages in a different theoretical framework. I treat the color and luminance as means of carrying orientation, rather than as an integral part of that orientation. My main question is whether orientation is coded in a different map for every color, and a different map for color and luminance, or whether it is coded in an abstract map independent of color and luminance (see Figure 3 a and b). The results may shed new light not only on how color and orientation are used and relate to one another, but may offer a more theoretically parsimonious interpretation of previously troublesome results. Interference tasks: Finding orientations in multiple colors In order to answer these questions about the interactions of orientation and color in human vision, I will be measuring the degree to which orientations interfere with each other when presented to the subject in different colors. These colors will be of the same apparent brightness, and of equal apparent brightness to the background on which they are presented. The degree of common representation can be inferred from the combined pattern of interference obtained from subjects performance in Stroop and Garner interference tasks. Stroop interference (Stroop, 1935) refers to the difficulty of identifying the hue of ink a word is printed in, if the word itself names a color. If the word has no relation to color (e.g., \"any\", \"more\", \"nothing\", etc., printed in red ink) subjects RTs do not differ from those obtained when of naming the color of a plain ink patch. If, however, the word names a color different from the hue of the ink in which it is printed (e.g., the word \"blue\" printed in red ink), color naming latencies and errors increase dramatically (see top half of Figure 4). Stroop interference in the ink naming task (Stroop, 1935) occurs when confusion between two conflicting messages increases reaction times or error rates. This interference suggests that the messages (color names) carried by both media (ink and the printed word) are concurrently available in an abstract medium-independent representation. Since both messages are task relevant a recovery of representational medium, either ink color or lexical code, may be necessary to make a correct classification. Insert Figure 4 about here When subjects must indicate the presence or absence of a vertical edge in a particular color among horizontal edges in the same color and vertical edges in a different color, a similar form of interference may occur (see bottom half of Figure 4). This interference between two vertical targets presented in different colors in a visual search task would suggest a representation for orientation that is independent of the color in which it is presented. A color-dependent representation of features like oriented edges would be consistent with the absence of Stroop interference. Another form of interference which may give converging evidence is Garner Orthogonal Set interference (Garner, 1974). Garner draws a distinction between two different ways stimulus dimensions may be combined. If the dimensions are \"integral\", manipulation of the differences along two dimensions result in a combined difference of near euclidean metric. That is, the combined difference is similar to that along the hypotenuse of a right triangle if manipulations of the two dimensions are treated as the two other sides of that same triangle. If the dimensions are \"separable\", manipulation of the differences along two dimensions result in a combined difference of near city-block metric. In this case, the dimension differences are additively, rather than exponentially related, and are therefore independent of one another. Subjects are presented with a classification task along one dimension, orientation for example, and stimuli are varied along a second orthogonal and task independent dimension (see Figure 5, note that circled stimuli represent those grouped in a particular classification task). For our purposes this second dimension will be the color in which the stimulus is represented. If the two dimensions are \"integral\" (as are Munsell value and chroma, see Garner 1974), this task is expected to be more difficult than the single dimension classification without orthogonal variation. If the dimensions manipulated are \"separable\" the increased but irrelevant variability would not effect task difficulty or reaction time since differences on the relevant dimension alone are sufficient to classify all of the stimuli. Insert Figure 5 about here This kind of interference pattern could be expected in the proposed experiments if orientation was represented separately in a series of color-dependent maps. A quick search would be required through as many maps as there are colors present in order to correctly find a vertical target. On the other hand, a lack of interference would be expected if orientation were coded in an abstract, color-independent representation, since variation of the color would have little or no effect on the abstract representation, and therefore on classification speeds. In summary, search through a series of color-dependent representations of orientation should cause no Stroop interference but should cause Garner interference. Conversely, search in a color-independent abstract representation for orientation should show Stroop interference, but no Gamer interference in the tasks described. Previous related research O'Connell and Treisman (1991) examined these patterns of interference between obliquely oriented lines, edges, and dot pairs of the same contrast in analogous conditions, as did Enns and Wig (1989) between orientations defined by contrast edges, lines and textures. Both sets of results are consonant with an abstract representation of orientation. O'Connell and Treisman (1991) found that subjects could search for a unique orientation defined by edges and lines. This was also true for virtual lines formed by dot pairs of the same luminance contrast relative to the background. If one of the dots in a pair was the brighter than the background and the other darker, subjects could no longer form a virtual line between them and could not search quickly for their orientation. This luminance-contrast restriction also held for oriented dot pairs of either the same or of different colors, relative to the background. If, however, search for lines, edges, or unicontrast dot pairs required that targets be distinguished from the same orientation carried by one of these two remaining media, search became slow and serial. These conditions are analogous to the Stroop interference task outlined above, and indicate a common orientation map for these media. These authors did not specifically test the interaction between colors of equal apparent brightness and orientation. They speculated that color representations, and color and luminance representations might be separate on the basis of physiological evidence presented by Livingstone and Hubel (1984, in O'Connell & Treisman, 1991). They also argued that such an organization would prevent subjects from forming virtual lines from bicontrast dot pairs, even those sharing a uniform color, and that this failure might be at the root of their inability to obtain parallel search for orientation defined by such stimuli. In the studies presented by Enns and Wig (1989), when the same target orientation was represented by both contrast and texture edges in the same display (e.g., a vertical luminance edge target among horizontal luminance edge and vertical texture edge distractors), Stroop interference occurred between the two instances, suggesting a common representation. Similarly, search for a single vertical target defined by either a contrast or a texture edge against a background of horizontal distractors defined by both contrast and texture edges failed to display the Garner interference expected of separate, edge type-dependent representations. With one ambiguous exception, our data support a single abstract representation for achromatic orientation. This research on achromatic (i.e., luminance\/texture) edges has raised similar questions about color edges. If edge orientation is represented independently of achromatic texture and luminance, it may also be represented independently of color. Evidence has shown that distinguishable orientation cues can be constructed from equiluminant color contrast borders, and that these orientations can be detected in parallel across the visual field (Cavanagh, et al. 1990). Such detection would be consistent with evidence of parallel vision stream research by Livingstone and Hubel (1988), and Shapley (1990). According to these researchers, chromatic and achromatic information is processed in parallel through different visual pathways from the retina through the Magno- and Parvocellular layers of the Lateral Geniculate Nucleus to different layers in VI . The Magnocellular pathway appears to convey primarily high temporal resolution, low spatial resolution, achromatic image information from the retina to layer 4Ca in VI , while the Parvocellular pathway appears to carry low temporal resolution, high spatial resolution, chromatic information to layer 4C(3 in VI . Here however, the distinction becomes less clear and indeed some functional combination takes place. The interblob areas in layers 2 and 3 of VI contain individual cells which respond to the orientation of an edge, whether its representational components be luminance, or equiluminant color contrasts, but remain insensitive to the specific colors or directions of contrast present. Cells in the blob areas of the same layers remain responsive to specific colors and brightness levels, but are not selectively responsive to orientations. At this early point in the visual processing stream we see the possible beginnings of an \"abstract\" representation for orientation. This physiological evidence is at odds with the medium dependent coding of orientation implicit in Treisman's (1988) multiple map theory. If the visual system can code for orientation independently of its medium, using cells in the interblob regions of VI , this abstraction of orientation from multiple media is available much earlier in the processing stream than posited by Treisman's multiple map theory. Use of this orientation abstraction capacity this early would drastically simplify the expanding plethora of orientation maps that Treisman proposes, while leaving medium information available (in the blob regions, and their connections to V2 and beyond) for more careful detailed and slower analyses, or parallel analyses for discrepancies on different criteria. The question of whether orientation can be abstracted from multiple colors in the same way it appears to be for achromatic texture and luminance edges can be directly addressed by testing for the presence of Stroop interference and the absence of Garner interference, as described above, for oriented bars of different equal-brightness hues presented against a neutral grey background of the same apparent brightness. On the basis of previous results from luminance and texture contrast borders, and apparent trends in neural processing of visual information in layers 2 and 3 in VI I can hypothesize a common representation of orientation independent of color. This will result in Stroop interference between representations of the same orientation by different colors of the same apparent brightness, and the absence of Garner interference in search for orientation when color is varied orthogonally. Experiment 1 Method Design. Seven conditions were run, with each subject participating in each condition in a randomized order. These seven conditions were broken up as follows: The first six conditions will constitute the baseline (1, 2), consistent (3, 4), and inconsistent (5, 6) conditions respectively for the Stroop interference task. Target present and absent search slopes will be analyzed in a 2 x 3 x 2 (Color x Distractor set x Target) three way, within subjects factorial design. Conditions 1, 2: Two baseline conditions in which the subject will search for a vertical target among horizontal distractors of the same color. This baseline task will be performed once in each color (see Figure 6a). 3, 4: Two distractor consistent conditions in which subjects will search for a vertical target of one color among horizontal distractors randomly selected from both stimulus colors. For example, subjects will search for a single green vertical bar among a random distribution of red and green horizontal bars (see Figure 6b). 5, 6: Two distractor inconsistent conditions in which subjects will search for a vertical target of one color among horizontal distractors of the same color and vertical distractors of the other color. For example, subjects will search for a single green vertical bar among green horizontal bars and red vertical bars (see Figure 6c). 7: A single condition in which subjects will search for a single vertical bar of either color among horizontal distractors randomly chosen from both colors. The order of target color will vary randomly in this condition. This condition is similar to the distractor consistent condition outlined in conditions three and four above, with the addition of target color uncertainty (see again Figure 6b). The first two and the seventh conditions (1, 2, 7) constitute the Garner interference task. Since the color baseline conditions (1, 2) should not differ, they will be combined and target present and absent search slopes will be analyzed with those from the last condition in a 2 x 2 (Distractor set x Target) two way within subjects factorial design. If the Stroop analysis described above finds significant differences between the two color baseline conditions, they will not be combined, but will be analyzed as separate levels of the Distractor set factor, resulting in a 3 x 2 (Distractor set x Target) two way within subjects factorial design. Insert Figure 6 about here Subjects. The subjects were nine University of British Columbia Students and Faculty, and one non-student, between the ages of 20 and 30. Five were experienced subjects from within Dr. Enns' visual perception lab at UBC, the remaining five were recruited from acquaintances of the experimenter and undergraduate volunteers who participated for experimental credit. Only subjects with normal or corrected to normal visual acuity and without congenital color defects were used. Visual acuity was assessed using a Bausch and Lomb Orthorater, and only subjects scoring greater than 6 (20\/29 Snellen equivalent) on the binocular near acuity plate were allowed to continue. Subjects misreading fewer than four plates in the 24 plate Ishihara set (Ishihara, 1972; Lakowski, 1969) under standard C L E . illuminant \"C\" were considered normal. Stimuli. The stimuli were horizontally and vertically oriented rectangles of red and green hues of the same apparent brightness, subtending visual angles of 1.1\u00b0 x 0.4\u00b0 at a viewing distance of 60 cm. One, six or twelve of these bar stimuli were presented on a grey field subtending a visual angle of 15.9\u00b0 horizontally, and 10.7\u00b0 vertically, in random positions on a grid of six equally spaced columns and four equally spaced rows. Within this grid, items were randomly jittered by up to 0.6\u00b0 vertically and up to 1\u00b0 horizontally to avoid target detection by simple broken symmetry or virtual texture irregularity. The density of items was controlled by placing them in a randomly placed sub-grid containing twice as many cells as there were items to be displayed. Preliminary data based on two subjects suggested that though individual differences in equal-brightness color choices are to be expected, these differences may in practice be quite small (see Table 1 below). Typical C L E . color values and color luminances for one subject, measured with a Minolta model CL-ICK) chroma meter and Minolta model LS-110 luminance meter respectively under normal viewing conditions, were as follows: Red x = .440, y = .297, luminance = 39.77 cd\/m 2 ; Green x = .297, y = .577, luminance = 40.55 cd\/m 2 ; Grey x = .281, y = .285, luminance = 39.19 cd\/m 2 . Note that the log difference between the most disparate physical luminances is on the order of 0.01 (log(40.55) - log (39.19) = 0.0148). Log differences of approximately 0.1 are the minimum that can be noticed by normal observers (R. Lakowski, personal communication, March 1, 1991). ) Insert Table 1 about here The stimuli were constructed using Photon Paint\u2122 for the Macintosh, and imported and displayed experimentally using IMaker 3.1\u2122 and VSearch\/Colour 3.1\u2122 (Enns, Ochs, & Rensink, 1990). Procedure. Subjects participated in three sessions: the first of roughly 90 minutes, then two of roughly 30 minutes each. The first session involved assessing each subject's visual acuity, color vision, and choosing red and green hues of the same apparent brightness for that subject. Barring excessive consumption of alcohol or caffiene, these values were expected to remain constant for each subject throughout the projected testing period. Changes in color sensitivity as a function of hormonal and light adaptation variations, or of physical changes in screen phosphors over extended periods, were not expected to interfere with subjects responses over the testing period. Different stimuli were constructed for each subject on the basis of their color choices, with the restriction that the neutral grey background remained the same, and that saturation and luminance for red and green remained at least 50% of maximum on the Macintosh Hue, Saturation, and Brightness (HSB) indices. These indices are arbitrary program values dividing the computer-controlled ranges of hue, saturation and brightness into 65535 possible steps each. This last restriction was necessary because the color red often had to be de-saturated to match the grey and green colors for brightness. The following two sessions consisted of running the seven experimental conditions in random order. Within each condition, subjects were shown the target and distractors, and were told to report the presence or absence of the target in each display by pressing keyboard keys as quickly as possible while keeping errors below 10%. The hand with which a given subject made the target present response in all conditions was varied randomly from subject to subject to avoid hand preference biases in reaction time. Color equi-brightness determination. Color equi-brightness on an AppleColor RGB monitor was determined by subjects response to a minimum flicker task. Subjects matched red and green separately to a predetermined uniform grey, by responding to the question \"does this flicker?\" while the experimenter adjusted color brightness to minimize the flicker. Two colors were considered to be the same apparent brightness when they fused into a uniform stable color without apparent flicker. Subjects initially made ten consecutive judgments of flicker as the brightness of one color was increased or decreased between pre-determined end points. The brightness levels immediately bounding the non-flickering values were made the end-points for another series of judgements using intervals of half the previous series size. Brightness was alternately increased and decreased in subsequent series. This procedure continued until three consecutive values were judged to fuse regardless of the direction of brightness change. The median brightness value for these fused intervals was then assessed for flicker, and if fused, was adopted as being of the same apparent brightness as the grey background. Subjects then compared the chosen reds and greens. If the colors appeared to flicker, the experimenter fine-tuned the subjects color choice by modifying the brightness of the green hue using the same criteria outlined above. If the resulting green hue did not fall outside the established fusion bounds for the grey background, the red and the modified green were adopted as being of the same apparent brightness as each other and as the neutral grey background. Preliminary tests compared color choices made when flickering the full screen to those made when only patches the size of the proposed stimuli were flickered. This test was suggested by the potential interaction between illuminant area and color in determining equal apparent brightness by the flicker method. Two subjects selected colors using the above method, in each of two randomly ordered stimulus conditions. For one selection subjects performed the protocol viewing the full screen, and for the other selection viewed the same screen through a mask cut from paper approximately the same color as the background grey. This mask allowed patches of color to be seen through apertures of the same size and position as stimuli on a standard twelve-item target present trial. Both the actual colors chosen, and the range around them which was judged not to flicker were compared across these two display conditions. All colors are cited in the Macintosh color wheel 0 hue\/saturation\/brightness (HSB) indices. Table 1 and Figure 7 present the Macintosh color palette values for the grey background and for the colors judged to be the same apparent brightness, as a function of display condition (see Table 1, Figure 7). The background grey used for all subjects was H=0, S=0, B=30000. All values for the color green refer to the 'B' component of the HSB indices, with H=21845 and B=65535. Similarly, all values for the color red refer to the 'S' component of the HSB indices, with H=0 and B = 65535. Insert Figure 7 about here First of all, the ranges of brightness judged to be the same apparent brightness were much narrower for the full screen display than for the narrow aperture, for both colors and both subjects. Secondly, the narrower equal brightness \"band\" for full screen choices fell well within the wider equal brightness range for small visual angle stimuli in all but one instance. This evidence from preliminary tests suggests that using the full screen flickering display not only resulted in color choices acceptable for patches the size of the proposed stimuli, but does so with greater precision than when smaller color patches are used. This full screen procedure was used when choosing colors of the same apparent brightness for subjects in these experiments. Results Stroop task. Mean response latencies and percentage errors were calculated for each subject in each of the Stroop interference conditions. A trial was discarded if the response of the subject was incorrect, or made after the 3 second response deadline. Mean increase in reaction time as a function of increased display size (search slope) was calculated for each condition using these reaction times. Figure 8 depicts RTs and errors as a function of target color, Stroop interference condition, target presence and display size. The RTs were similar in most conditions, and with the exception of the distractor inconsistent conditions, where RTs increased as a function of display size, error rates remained below 10%. These patterns suggest that no speed accuracy trade off influenced subjects responses. The RT increase in the inconsistent conditions suggests a serial search rate different from the baseline and consistent conditions, which is indicative of a Stroop interference effect. Insert Figure 8 about here Four-way within-subjects Analyses of Variance were conducted on subject RTs and error rates for the Stroop interference conditions. They examined the factors of medium, condition, target presence and display size. The analysis showed no four-way interaction, F(4,36) = .28, p > .05, indicating that the significant three-way interaction (F(4,36) = 3.69, p < .05) of condition, target condition, and sisplay size followed the same pattern whether the target was green or red. The analysis of error rates for these conditions also failed to find a four-way interaction between Media, Condition, Target presence, and Display size (F(4,36) = 1.90, p > .05). The significant three-way interaction between Condition, Target presence, and Display size indicates that error rates remained low and constant for the baseline and consistent conditions, regardless of target color, and increased considerably as a function of display size in the inconsistent distractor conditions. These changes parallel similar increases in reaction time documented above, and therefore argue against RT contamination by speed-accuracy trade off effects. This argument is enhanced by a significant positive correlation between mean error rates and RTs across all conditions (r = .42, p < .05). Figure 9 depicts search slopes as a function of target color, condition, and target presence. The similarity of the slope functions for the different target colors suggests that any interference between vertical targets and distractors of different colors and orientations is symmetrical. The small search slopes for the baseline and consistent conditions are consistent with fast parallel search, while the larger slopes for the inconsistent conditions indicate slower serial search indicative of Stroop interference. Insert Figure 9 about here A three-way within-subjects Analysis of Variance was computed on mean search slopes derived from these RT data, examining the factors of medium, distractor condition and target presence. The three-way interaction failed to reach significance (F(2,18)=. 12, p > .05), indicating that the two-way condition by target-presence interaction (F(2,18) = 7.21, p < .01) was the same regardless of target color. Subsequent simple main effects analysis showed significant increases in both target present and target absent search slopes as a function of condition (F(2,18) = 41.21, p < .001; and F(2,18) = 48.96, p < .001 respectively). Individual comparisons using the Tukey HSD method ((Sx)(.95a6,18) = 4.62, and (Sx)(.99g6,18) = 5.77 for cc=.05 and .01 respectively) showed no significant differences between baseline and consistent slopes, but found inconsistent present and absent slopes significantly greater than both the baseline and consistent condition slopes. These results show a strong Stroop interference effect in the inconsistent condition compared to the baseline orientation search task. This interference cannot be attributed to the presence of different color distractors alone, since the consistent condition was not found different from the baseline condition in any respect. Nor can this Stroop interference be attributed to speed-accuracy trade off, since errors remained effectively constant for all conditions and increased only when RTs increased in the inconsistent condition. Garner task. Figure 10 depicts subject RTs and error rates as a function of target condition, display size, and target presence. Subjects reaction times show similar patterns in each condition, with absent RTs greater than present conditions, the greatest difference being at display sizes of one. As display sizes increase, RTs appear to increase slightly, and differences between present and absent slopes appear to decrease. Errors follow a different pattern. Though all error rates are well below 10%, target absent errors remain lower than present errors and appear to decrease markedly as a function of increasing display size. Target present errors appear to remain roughly constant, increasing the difference between present and absent error rates as a function of increasing display size. These results suggest a mild speed-accuracy trade off, which if rectified would equalize RTs for all conditions. The only apparent difference not accounted for by this trade off is the greater target absent RTs across all target conditions. Insert Figure 10 about here Three-way within-subjects Analyses of Variance were conducted on subjects RTs and error rates. They examined the factors of condition, target presence and display size. The three-way interaction in the RT analysis failed to reach significance (F(2,36)=.63, p > .05), indicating that the significant interaction between target presence and display size (F(2,18) = 6.9, p < .01) was not meaningfully different from one condition to another. Target absent RTs remained roughly constant, while target present RTs increased linearly as a function of increasing display size. These patterns are discussed statistically in the slope analysis below. Analysis of subjects error rates revealed no significant three-way interaction (F(4,36) = .01, p > .05), indicating that the significant two-way interaction between target presence and display size remained consistent over conditions (F(2,18) = 10.01, p < .01). As apparent in Figure 10, target present error rates remained roughly constant, while target absent errors decreased linearly as a function of increasing display size. Though all error rates remained below 10%, the decrease in error rates in conditions with similar RTs suggests a mild speed accuracy trade off as a function of display size. The presence of such a trade-off is suggested by a moderate, nonsignificant negative correlation between subject error rates and RTs across all conditions (r = -.36, p > .05). If RTs are increased to equalize error rates across conditions, both present and absent RTs for all conditions would rise gradually as a function of display size. The magnitude of this speed-accuracy trade off does not appear to vary across conditions, suggesting that the baseline red and green target conditions are not systematically different from the alternating green\/red vertical target condition in any way. Figure 11 depicts mean search slopes for all subjects as a function of target condition and target presence. Search slopes appear constant, well under 10 ms\/item across all conditions. No systematic difference in search slopes appears as a function of condition, and all conditions exhibit search slopes consistent with fast parallel search for the target independent of display size. Insert Figure 11 about here A two-way within-subjects Analysis of Variance was conducted on these means. Neither the two-way interaction, nor either of the main effects achieved significance (F(2,18)=.63, p > .05; F(l,9) = 3.76, p > .05; and F(2,18) = 1.7, p > .05 for the interaction, target presence, and target condition effects respectively). Though the RT data above showed a significant increase in RT as a function of display size, even after adjustment to compensate for the speed-accuracy trade off, these RT increases were consistent and remarkably small across all tested conditions: well under 10 ms\/item. These search rates are consistent with fast parallel search for the target independent of display size or target color. Discussion In this first experiment, subjects were able to conduct a fast parallel search for orientation defined by red or green color bars of the same apparent brightness against a grey background of the same brightness. The presence of consistent orientation distractors of either color did not interfere with this search in any way. When the distractor set included bars of the same orientation as the target, but of a different color, a strong symmetrical Stroop interference was documented. No Garner interference occurred when subjects searched for a vertical bar of randomly alternating color among mixed color, uniform orientation distractors. Combined, these two test results argue for a single, color independent representation for orientation. Experiment 2 Existing results (Enns & Wig, 1989), and data from Experiment 1 suggest a common orientation map for luminance and texture, and a common map for different colors. This raises the possibility of an abstract orientation map for color and luminance. Evidence presented above for an abstract orientation map for orientation defined by different colors of the same apparent brightness tests only one part of Treisman's multiple map theory. So far in this thesis, comparisons have been made only between different levels of color within what Treisman would describe as the color module (See Figure 1). Though troublesome for the multiple map theory, interference within only one module does not constitute a strong test and disconfirmation. A stronger test would examine interference across media from different proposed modules. One test proposed by the physiological data outlined in the introduction would be a comparison of color edges to luminance edges. Not only are these two modules held separate in the classic feature integration theory, and still held separate in Treisman's most recent work (O'Connell & Treisman, 1991), but the information necessary to complete such a search task is carried separately by the Magno- and Parvocellular pathways at least as far as VI . Physiological evidence discussed above describes single cells in the interblob regions of layers 2 and 3 of VI which detect edges of a particular orientation independent of their color or luminance composition, and of the relative brightnesses of the contrasting edge halves. An examination of the relation between oriented luminance and color edges would complete the interactions of luminance and texture edges begun by Enns and Wig (1989), and the continued investigation of color edges documented above, to present a unified body of evidence for a single medium-independent orientation map for visual search. Behavioral evidence for a common representation of orientation across color and luminance defined orientation edges would seriously question the multiple medium-dependant orientation map hypothesis and show an abstraction of orientation at a much earlier stage of visual processing than previously posited. This experiment will assess the interactions between color and luminance media in visual search for orientation. A single representation for orientation is hypothesized, which will produce Stroop interference between luminance and color defined edges of the same orientation. No Garner interference should occur. Method The design and procedure was the same for this experiment as for Experiment 1 above. All but one of the same subjects from Experiment 1 were used. One new volunteer was recruited subject to the requirements specified in Experiment 1. The stimuli and their presentation duplicated those used in Experiment 1, with the exception that white oriented rectangles were substituted for green bars in all tests. Results Stroop task. As in Experiment 1, mean RTs and error rates were calculated for each subject and condition after discarding incorrect trials, and mean search slopes were calculated from the remaining RTs. Figure 12 depicts RTs and error rates as a function of target color\/luminance, Stroop interference condition, target presence and display size. RTs were similar in most conditions, and error rates remained below 10% for all conditions. Errors approached 10% only in the distractor inconsistent condition which showed a corresponding increase in RT as a function of increasing display size. This pattern is better described as speed-accuracy correspondence than as speed-accuracy trade off, and suggests that the inconsistent distractor task became more difficult as display size increased. The RT increase in the distractor inconsistent condition suggests a serial search rate different from the baseline and consistent distractor conditions, which indicatives a Stroop interference effect. Insert Figure 12 about here Four-way within-subjects Analyses of Variance were conducted on subject RTs and error rates, and examined the factors of medium, condition, target presence and display size in both cases. RTs showed no four-way interaction (F(4,36) = 2.58, p > .05) indicating that the significant three-way interaction of condition, target presence and display size (F(4,36) = 4.02, p < .01) followed the same pattern whether the target was red or white. The analysis of error rates for these conditions failed to find a four-way interaction (F(4,36) = 0.69, p > .05). The highest order interaction to reach significance was the two-way interaction between target condition and size (F(4,36) = 7.06, p < .01), indicating that while errors decreased as a function of display size for the baseline and consistent conditions, they increased in the inconsistent condition. These patterns argue against the contamination of these results by a speed-accuracy trade off. This argument is strengthened by the presence of a significant positive correlation between RTs and error rates across all conditions (r = .47, p < .05). Figure 13 depicts search slopes as a function of target color\/luminance, condition, and target presence. The slightly different pattern of slopes for the two target colors suggests that the effects of vertical distractors of one color on search for vertical targets of another may not be perfectly symmetrical. The great similarity of general pattern and direction of slope change in these conditions, however, argues for a difference in magnitude rather than in type or quality of interference. In both cases, the small search slopes for the baseline and consistent distractor conditions is consistent with fast parallel search while the larger slopes for the inconsistent conditions indicate slower serial search consistent with Stroop interference. Insert Figure 13 about here A three-way within-subjects Analysis of Variance, examining the factors of medium, condition and target presence, was computed on mean search slopes derived from the RT data analyzed above. The three-way interaction attained significance (F(2,18) = 6.43, p < .01) necessitating separate two-way analyses to examine the effects of target color on Stroop interference. Two-way within-subjects Analyses of Variance were conducted to examine the factors of condition and target presence in the white-target and red-target conditions respectively. The white-target analyses yielded a significant effect of condition (F(2,18) = 43.24, p < .0001). Subsequent individual comparisons between baseline, consistent and inconsistent slopes averaged across target presence, utilizing the Tukey HSD method, ((Sx)(.95q3,18) = 4.81, and (Sx)(.99\u00a33,18) = 6.26 for a=.05 and .01 respectively) showed no significant differences between the mean baseline and consistent search slopes, and significant differences between both of these and the inconsistent condition slope. These results show a strong Stroop interference effect in the inconsistent condition compared to the baseline and consistent search conditions. This interference cannot be attributed to the presence of different color\/luminance distractor alone, since the consistent condition was not found different from the baseline condition in any respect. The red-target analysis showed a significant two-way interaction between distractor condition and target presence (F(2,18) = 12.87, p < .001). Subsequent Tukey HSD individual comparisons ((Sx)(,95g6,18) = 3.45, and (Sx)(.99a6,18) = 4.30 for a=.05 and 01 respectively) showed no significant differences between baseline and consistent slopes, but found inconsistent present and absent slopes significantly greater than both the respective baseline and consistent condition slopes. The significant interaction seems to stem from the relation between the present and absent search slopes in each condition. Though the absent slopes appear lower than the present slopes in both the baseline and consistent conditions, no significant differences were found. Conversely, the absent slope is significantly greater than the present slope in the inconsistent condition. This pattern of results resembles that of the white target analysis above in all details save this last difference between present and absent slopes in the inconsistent condition. In absolute terms, a strong Stroop interference is apparent in the inconsistent condition relative to the baseline condition. Again, this effect cannot be attributed to the presence of different color distractors alone, since no differences were found between baseline and consistent condition slopes. The relative differences in slope patterns is small, since compared across target color condition, all slopes are of similar magnitude and share a nearly identical pattern of interference. If this difference in pattern indicates an interference asymmetry with white vertical distractors causing greater interference with search for red vertical targets than vice-versa, this asymmetry is small enough to allow significant Stroop interference of approximately the same magnitude in both directions. Garner task. Figure 14 depicts subjects RTs and error rates as a function of target condition, display size and target presence. Subject RTs show similar patterns for all three conditions, with target absent RTs greater than target present RTs, and the greatest difference being at the smallest display size. RTs appear to increase slightly as a function of display size, and differences between present and absent RTs appear to decrease. Error patterns, though similar across conditions, follow a pattern quite different from subject RTs. Target absent error rates are slightly lower than target present, and the difference appears to grow as both decrease as a function of increasing display size. This pattern suggests a mild speed accuracy trade-off which, if compensated for, would equalize RTs across all display sizes. Insert Figure 14 about here Three-way within-subjects Analyses of Variance were conducted on subjects RTs and error rates, to examine the factors of condition, target presence and display size. The three-way interaction in the RT analysis failed to reach significance (F(4,36) = 1.05, p > .05) indicating that the significant two-way interaction between target presence and display size (F(2,18) = 5.35, p < .05) held for all target conditions. Target absent RTs remained roughly constant while target present RTs increased linearly as a function of display size. These results will be statistically examined in the slope analyses below. Analyses of subject error rates revealed only a significant effect for display size (F(2,18) = 4.76, p < .05), and no mitigating interactions or other effects. Error rates appeared to decrease uniformly as a function of increased display size. Though all error rates remained below 10%, this significant decrease combined with an increase in subject RTs over the same display size conditions indicates a speed-accuracy trade off. Further evidence of this trade off is provided by a significant negative correlation between between RTs and error rates across all conditions (r = -.55, p < .05). If RTs are increased to reduce error rates in the smallest display size conditions and thereby equate errors over all conditions, RTs would remain constant or diminish as a function of increasing display size. This speed accuracy trade-off does not appear to vary across conditions. No reason exists to suggest that the relative differences between alternating white\/red target conditions would be affected in any way. Figure 15 depicts mean search slopes for all subjects as a function of target condition and target presence. Search slopes appear roughly constant, well under 10 ms\/item in all conditions. No systematic difference in search slopes appears as a function of condition, and all conditions exhibit search slopes consistent with fast parallel search for the target independent of display size. Insert Figure 15 about here A two-way within-subjects Analysis of Variance was conducted on these mean slopes to examine the condition and target presence factors. Neither the two-way interaction, nor the main effects of either target presence or condition attained significance (F(2,18) = 1.25, F(l,9) = 3.76, F(2,18) = 0.89, respectively. All p > .05). Though no support can be given by acceptance of the null hypothesis, the fact that all conditions displayed slopes well under 10 ms\/item argues strongly for the absence of any Garner interference when subjects search for a vertical edge of randomly chosen color\/luminance when compared to each baseline condition. Search rates and arguably search strategies appeared the same for all conditions: Fast parallel search independent of display size. Discussion In this second experiment, subjects were able to conduct a fast parallel search for orientation defined by uniform red color or white luminance bars, against a grey background. The presence of consistent orientation distractors of both media did not interfere with this search in any way. When the distractor set included bars of the same orientation as the target, but of a different medium, a strong Stroop interference was documented. Some asymmetry is apparent, but the difference occurs only in the relation between present and absent slopes in the inconsistent distractor conditions, and is of very small magnitude compared to the main Stroop interference effect. The interference is functionally, if not statistically, symmetrical. No Garner interference occurred when subjects searched for a vertical bar of randomly alternating color and luminance among mixed color and luminance uniform-orientation distractors. Combined, these two test results suggest that visual search is based on a single abstract representation of orientation (i.e., one that independent of both color and luminance). General Discussion The series of medium-specific orientation maps posited by Treisman (1988) are untenable in light of the present results. All results presented here and in previous studies in collaboration with Enns (Enns & Wig, 1989) point to a single set of medium-independent orientation maps. These results are consistent with those presented by O'Connell and Treisman (1991), but also extend them to include interactions between colors of the same apparent brightness and between color and luminance. In addition to better explaining current data, these abstract maps are simpler than the series of special case instances posited by Treisman (1988). This research presents evidence for an orientation representation accessible to visual search that is independent of color or luminance definition. Both effectively convey orientation information in isolation, and convey the same orientation when presented together. This causes Stroop interference in visual search tasks whenever the same orientation is present in both the target and distractor sets in different media. Similarly, abstraction of orientation is possible from red, green and luminance media simultaneously and without cost, as evidenced by the lack of Garner like interference documented in both experiments. When these results are combined with those presented by O'Connell and Treisman (1991) and Enns and Wig (1989), they suggest a common representation of orientation for edges conveyed by unicontrast dot pairs, luminance, texture, and edges formed by colors of the same apparent brightness. Though not yet tested, interactions similar to these are predicted for color and texture edges. This last abstraction of orientation from texture, color and luminance edges is particularly interesting, since each of these surface characteristics is posited by many to be separately and independently coded by different pathways in the precorneal and striate visual systems (Cavanagh, et. al., 1990; Lennie, Trevarthen, Van Essen & Wassel, 1990; Livingstone 6k Hubel, 1988; Shapley, 1990; Treisman, Cavanagh, Fischer, Ramachandran & von der Heydt, 1990; Zrenner, Abramov, Akita, Cowey, Livingstone & Valberg, 1990). Chromatic information is handled primarily by the parvocellular system, while the magnocellular system is sensitive to motion and fine gradations in achromatic contrast. However, orientation is coded by the Parvo- and Magnocellular systems in VI (Livingstone & Hubel, 1988). Specifically, cells in the interblob areas in layers 2 and 3 of VI respond to orientation independent of edge color or luminance contrast sign. These cells could then provide a means of coding orientation abstractly across two media > putatively considered functionally discrete. Behavioral evidence from Experiments 1 and 2 is consistent with this account, and substantiates the more general claim for a more complex early visual processing than previously conceived. Stroop interference across these media suggests a much more complex visual processing than previously posited for the visual search task. Luminance and color contrasts can be measured by comparing any two points in an image, and have therefore been called first-order characteristics (Cavanagh, et al., 1990). A texture is defined by a relation between one or more of such first order characteristics, and therefore requires at least a pair of two-point comparisons to define a single texture. The boundary between two textures can only be determined by a change in the relations between these denning second-order characteristics. This texture boundary could therefore be considered a third-order characteristic. Since texture defined oriented edges require such vastly more complex visual processing, well beyond that demonstrated within VI , it is very surprising to find a common representation for orientation for texture and luminance based orientation (Enns & Wig, 1989). Such a finding defies both the functional limits of visual search implicit in Treisman's Feature Integration Theory, and known physiological limits within VI (Livingstone & Hubel, 1988). Abstract orientation representation for edges and virtual lines generated by unicontrast dot pairs documented by O'Connell and Treisman (1991) also challenges the relegation of visual search decisions to processes in the striate visual cortex. Further evidence for the likelihood of visual search to be based on representations beyond the primary visual cortex comes from results published by Enns (1991) and Enns and Rensink (1990). These researchers documented parallel search for three-dimensional orientation, direction of lighting, curvature from shading, and line drawings of orthogonal polyhedra. All of these results indicate that the visual system not only codes simple surface features such as luminance, color, and the orientation of luminance and color borders, but appears to develop codes for each orientation independent of the color, luminance, contrast, or texture of that edge. When considerations of the complexity of the processing necessary to obtain such an abstract representation are combined with evidence cited for parallel search for three-dimensional and scene-based properties, we must consider visual search to use representations much more complex than previously posited. References Beck, J. , Sutter, A., & Ivry, R. (1987). Spatial frequency channels and perceptual grouping. Computer Vision. Graphics, and Image Processing. 37. 299-325. Cavanagh, P., Arguin, M., & Treisman, A. (1990). Effect of surface medium on visual search for orientation and size features. Journal of Experimental Psychology: Human Perception and Performance. 16(3). 510-547. Enns, J . T. (in press). The nature of selectivity in early human vision. In B. Burns (Ed.), Percepts, concepts, and categories: The representation and processing of information. Amsterdam: Elevier Science. Enns, J . T., Ochs, E. P., & Rensink, R. A. (1990). VSearch: Macintosh software for experiments in visual search. Behavior Research Methods. Instruments. & Computers. 22(2), 118-122. Enns, J . T., & Rensink, R. A. (1990). Influence of Scene-based properties on visual search. Science. 247. 721-723. Enns, J . T , & Wig, P. S. (1989). The message and the medium in early vision. Paper presented at the Psychonomic Society, Atlanta, Georgia. Garner, W. R. (1974). The Processing of Information and Structure. New York: Erlbaum. Ishihara, S. (1972). Tests for Colour Blindness. Tokyo: Kanehara Shuppan. Lakowski, R. (1969). Theory and practice of colour vision testing: A review. Part 2. British Journal of Industrial Medicine. 26. 265-288. Lennie, P., Trevarthen, C , Van Essen, D., & Wassel, H. (1990). Parallel processing of visual information. In L. Spillmann & J . S. Werner (Eds.), Visual perception: The neurophysiological foundations (103-128). London: Academic. Livingstone, M., 6k Hubel, D. (1988). Segregation of form, color, movement, and depth: Anatomy, physiology, and perception. Science. 240. 740-749. McLeod, P., Driver, J., 6k Crisp, J . (1988). Visual search for conjunctions of movement and form is parallel. Nature. 332. 154-155. Nakayama, K., 6k Silverman, G. H. (1986). Serial and parallel processing of visual feature conjunctions. Nature. 320. 264-265. O'Connell, K., 6k Treisman, A. (1991). Shared orientation coding for lines, dot pairs, and edges. Manuscript submitted for publication. Shapley, R. (1990). Visual sensitivity and parallel retinocortical channels. Annual Review of Psychology. 41. 635-658. 5 5 Stroop, J . R. (1935). Studies of interference in serial verbal reactions. Journal of Experimental Psychology. 18. 643-662. Treisman, A. (1986). Features and objects in visual processing. Scientific American. 255(5). 114b-125. Treisman, A. (1988). Features and objects: The Fourteenth Bartlett Memorial Lecture. The Quarterly Journal of Experimental Psychology. 40A(2). 201-237. Treisman, A. (In press). There's more to search than similarity: Conjoining features can take time. Treisman, A., Cavanagh, P., Fischer, B., Ramachandran, V. S., & von der Heydt, R. (1990). Form perception and attention: Striate cortex and beyond. In L. Spillmann & J . S. Werner (Eds.), Visual perception: The neurophysiological foundations (273-316). London: Academic. , Treisman, A., & Gelade, G. (1980). A feature-integration theory of attention. Cognitive Psychology. 12. 97-136. Treisman, A., & Gormican, S. (1988). Feature analysis in early vision: Evidence from search asymmetries. Psychological Review. 95(1), 15-48. Wolfe, J . M., Cave, K. R., & Franzel, S. L. (1989). Guided search: An alternative to the feature integration model for visual search. Journal of Experimental Psychology: Human Perception and Performance. 15. 419-433. Zrenner, E., Abramov, I., Akita, M., Cowey, A., Livingstone, M., 6k Valberg, A. (1990). Color perception: Retina to cortex. In L. Spillmann 6k J . S. Werner (Eds.), Visual perception: The neurophysiological foundations (163-204). London: Academic. Table 1 Color Brightness Levels and Fusion Ranges as a Function of Screen Masking Condition Mask No Mask Color Brt\/Sat. a Fused Brt\/Sat. Fused Grey 30000 30000 ---Subject 1 Red 47000 +\/- 3000 46105 +\/- 15 Green 40000 +\/- 4000 40750 +\/- 625 Subject 2 Red 45500 +\/- 500 45150 +\/- 50 Green 40400 +\/- 1000 40865 +\/- 250 aThe brightness of the colors green and grey, and the saturation of the color red are recorded in this table. Figure Captions Figure 1. Treisman's feature integration theory. Thin lines denote preattentive processes; heavy lines denote attentive processes as a result of focused attention in the master map of locations. Figure 2. Examples of five different surface media which may carry a vertical edge message (see text for details). Figure 3. Proposed organization of orientation maps for different surface media in Treisman's (1988) multiple map theory (a) and the proposed abstract map theories (b). Figure 4. Examples of original Stroop interference paradigm for color and words (a) and for color and orientation in Experiments 1 and 2 (b). Figure 5. Example of Garner interference paradigm for shape and size (top) and for color and orientation (bottom). Circled items represent target categories in the Garner classification task (see text for details). Figure 6. Example displays for the Baseline (a), Consistent (b), and Inconsistent (c), Stroop task conditions. Item (b) also serves as an example of the Garner task display for one target color. Figure 7. Macintosh brightness (Green) and saturation (Red) levels describing fusion ranges for equal-brightness color choices. Levels are plotted as a function of masking condition and subject (see text for details). Figure 8. Mean reaction times and error rates for the Stroop conditions in Experiment 1, plotted as a function of display size: Green Baseline (a), Green Consistent (b), Green Inconsistent (c), Red Baseline (d), Red Consistent (e), Red Inconsistent (f). Figure 9. Mean search slopes for the Stroop conditions in Experiment 1, plotted as a function of condition: Green target (a), Red target (b). Figure 10. Mean reaction times and error rates for the Garner conditions in Experiment 1, plotted as a function of display size: Green baseline (a), Red Baseline (b), Alternating Target (c). Figure 11. Mean search slopes for the Garner conditions in Experiment 1, plotted as a function of condition. Figure 12. Mean reaction times and error rates for the Stroop conditions in Experiment 2, plotted as a function of display size: White Baseline (a), White Consistent (b), White Inconsistent (c), Red Baseline (d), Red Consistent (e), Red Inconsistent (f). Figure 13. Mean search slopes for the Stroop conditions in Experiment 2, plotted as a function of condition: White target (a), Red target (b). Figure 14. Mean reaction times and error rates for the Garner conditions in Experiment 2, plotted as a function of display size: White baseline (a), Red Baseline (b), Alternating Target (c). Figure 15. Mean search slopes for the Garner conditions in Experiment 1, plotted as a function of condition. 61 Figure 1: Treisman's feature integration theory. Thin lines denote preattentive processes; heavy lines denote attentive processes as a result of focused attention in the master map of locations STIMULI SPOTLIGHT OF A T T E N T I O N ( 62 Figure 2: Examples of five different surface media which may carry a vertical edge message (see text for details). A'\/s\/s&A A A A A A ss\/x\/s A A'ts\/sftA A ?s\/s& A A A A ss\/s\/s A A 'Wj^S A A A A'\/s\/s&A ft y \u00a3| y S 'A' A '& A ft ft ft ft ft ft ft ft ft ft ft ,ft ft LUMINANCE TEXTURE COLOR BINOCULAR RELATIVE DISPARITY MOTION 63 Figure 3: Proposed organization of orientation maps for different surface media in Treisman's (1988) multiple map theory (a) and the proposed abstract map (b) LUMINANCE COLOUR ORIENTATION ORIENTATION B ABSTRACT ORIENTATION TEXTURE ORIENTATION 64 Figure 4: Example of the Stroop interference paradigm for color and words (top) and for color and orientation (bottom, see text for details on stroop tasks and conditions). A mmr\/\/\/\/%WA mmi BLUE BASELINE CONGRUENT INCONGRUENT B BASELINE TARGET DISTRACTOR K(red) CONSISTENT ;(red)\\ INCONSISTENT c Figure S: Example of Garner interference paradigm for shape and size (top) and for color and orientation (bottom). Circled items represent target categories in the Garner classification task (see text for details). L U \u2014 CO ORIENTATION 66 Figure 6: Example displays for the Baseline (a), Consistent (b) and Inconsistent (c) Stroop task conditions. Item (b) also serves as an example of the Garner task display for one target color. A B 0 :-:-:-:-:-:-:::-:kV^^ :-:-:-:-:-:-:-:-:-:-:-:-:-:-:-:-:-:-x-:-:-:-:-:->^ >v^ ti c m Vj::::::::::::::::::-::-::::::::::::-: m 67 Figure 7: macintosh brightness (Green) and saturation (Red) levels describing fusion ranges for equiluminant color choices. Levels are plotted as a function of masking condition and subject (see text for details). ^ 49000 H CO > o CO CO CO sz (0 o CO 39000 44000 - -L co CO > CD CO CO o CO \u2022 Subj 1 Mask H Subj 1 No Mask E3 Subj 2 Mask 0 Subj 2 No Mask Red Gren Color 68 Figure 8: Mean reaction times and error rates for the Stroop conditions in Exp; 1, plotted as a function of display size: Green Baseline (a), Green Consistent (b) Green Inconsistent (c), Red Baseline (d), Red Consistent (e) and Red Inconsis-tent (f). in o in \u2022 i \u2022 \u2022 \u2022!i \u2022 \u2022 \u2022 \u2022 i \u2022 \u2022 \u2022 \u2022i\u2022\u2022 \u2022 \u2022 i \u2022 > \u2022 \u2022 i\u2022 E23 r \u2022 \u2022 [ \u2022 \u2022 \u2022 \u2022 | \u2022 \u2022 \u2022 \u2022 | \u2022 \u2022 \u2022 \u2022 | \u2022 \u2022 \u2022 \u2022 | \u2022 Of Kl 'a. VI -i-> CO CO c o o o o o o o o o o o o in o in o in o in o in o in o in o o o o o o o o o o o i n o i n o i n o i n o i n (26) S J O J J 3 m o in CM _ o o o o o o o o o o o DQ in o in o in o in o i n o i n oo oo r- r-- vo vo m in T J - w oo co r-- r-- -sO \u2022JD in i n * * w (26) S J 0 J J 3 m o m E- ^ \u2022\u2022\u2022\u2022\u2022\u2022\u2022\u2022\u2022\u2022\u2022\u2022\u2022 Kl VI a. O O O O O O O O O O O m o in o in o in o in o in oo co r- r- >x> in m * * r o (26) S J O J J 3 m o m CO 3 I \u2022 \u2022 < \u2022 11 \u2022 \u2022 \u2022 I \u2022 I \u2022 \u2022 1111 \u2022 I \u2022 CN O O O O O O O O O O O in o i n o i n o i n o i n o i n oo co r-- vo M> in in t * ro - < ( S U J ) 3UJL^ uoi^oeay (26) SJ0JJ3 I'\"1!\"\"!\"\"!\"\"!\"\"!1'\"!\" o o o o o o o in o in o in o in co co r-- r- vo vo in m o m CN Kl a. O O O O o in o in 111 * * K ) ( S U J ) euu} uoi^oeay Green Target Red Target 30 25 -20 -15 10 -5 -0 -5H -10 O
o _ o CD 2 CD Baseline Consitent Inconsitent Baseline Consitent Inconsitent CD rt- CD CD \u00a32. 70 Figure 10: Mean reaction times and error rates for the Garner conditions in Experiment 1, plotted as a function of display size: Green target (a), Red target (b) and Alternating target (c).