University of Southern California Dissertations and Theses

Invariance to changes in contrast polarity in object and face recognition

(USC Thesis Other)

Invariance to changes in contrast polarity in object and face recognition

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content INVARIANCE TO CHANGES IN CONTRAST POLARITY IN OBJECT AND FACE RECOGNITION by Marissa Nederhouser A Thesis Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment o f the Requirements for the Degree MASTER OF ARTS (PSYCHOLOGY) May 2003 Copyright 2003 Marissa Nederhouser R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. UMI Number: 1416569 Copyright 2003 by Nederhouser, Marissa All rights reserved. UMI UMI Microform 1416569 Copyright 2003 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. UNIVERSITY OF SOUTHERN CALIFORNIA THE GRADUATE SCHOOL UNIVERSITY PARK LOS ANGELES, CALIFORNIA 90089-1695 This thesis, written by under the direction of h < ■ ■ thesis committee, and approved by all its members, has been presented to and accepted by the Director of Graduate and Professional Programs, in partial fulfillment of the requirements for the degree of MASTER OT ARTS Date May 16. 2003 Thesis Comprittee Chair T . ( ..I? R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. Dedication This thesis is dedicated to Cary, Denise, Veronica, and Mike, who probably have put in more hours worrying about this thesis than even I have... and kept telling me that staying in L.A. was a good idea. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. iii Acknowledgements I would like to acknowledge the efforts put forth by the following persons towards the completion o f the research presented in this thesis: My advisor Irving Biederman’s consistent supervision in the planning o f this line o f research, the development of the stimuli, and all his input into the analysis done. My thesis committee comprised of Bosco Tjan, Zhong-lin Lu, and Laura Baker’s wonderful comments and criticisms leading to better and brighter research and writing. Mike Mangini’s help with the initial code generating the stimuli and all o f his insight into the methodology and analysis of the data. Junmei Zhu and Kazunori Okada’s help with modifications made to the Gabor Jet Similarity assessment code. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. Table of Contents Dedication ii Acknowledgements iii List of Figures v Abstract vii Introduction: Reported Differences in Object and Face Recognition 1 1. Experiment 1: Novice and Expert Sequential Matching of Blobs Generated by the Rotational-Variation o f the Harmonics of a Sphere 5 1.2 Experiment 1: Methods 6 1.2.1 Exp. 1 Subjects 6 1.2.2 Exp. 1 Stimuli 7 1.2.3 Exp. 1 Design 11 1.2.4 Exp. 1 Procedure 14 1.3 Experiment 1: Results & Discussion 17 2. Experiment 2: Novice and Expert Forced-choice Match-to-sample o f Blobs Generated by the Amplitude-Variation of the Harmonics o f a Sphere 23 2.2 Experiment 1: Methods 24 2.2.1 Exp. 2 Subjects 24 2.2.2 Exp. 2 Stimuli 24 2.2.3 Exp. 2 Design 26 2.2.4 Exp. 2 Procedure 29 2.3 Experiment 2: Results & Discussion 31 General Discussion 36 References 43 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. List of Figures 1. Contrast Inverted Chairs and Faces 1 2. Results from Subramaniam & Biederman (1997) 2 3. Illustration o f the Lades et al. (1993) architecture 3 4. Shepard & Cermak’s (1973) procedure for generating a free form 6 5. Production of Rotationally-Varied Blob Stimuli 7 6. Assessing Similarity Values. 9 7. City Block Distances and Gabor Jet Similarity Values 10 8. Experimental Conditions for Exp. 1 13 9. Example “Same” Trial for Experiment 1 14 10. Example “Different” Trial for Experiment 1 16 11. GJS, RTs, and Error Rates (Figure 1 o f 2) 19 12. GJS, RTs, and Error Rates (Figure 2 of 2) 19 13. Exp. 1— RTs for Novices and Experts on Same vs. Different Contrast Trials 20 14. Exp. 1— Error Rates for Novices and Experts on Same vs. Different Contrast Trials 20 15. Exp. 1— RTs for Novices and Experts on Same vs. Different Shape Trials 22 16. Exp. 1— Error Rates for Novices and Experts on Same vs. Different Shape Trials 22 17. Exp. 2—Production of Amplitude-Varied Blob Stimuli 25 18. Exp. 2 — Stimuli (amplitude-varied blobs) 26 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 19. Exp. 2—RTs for Novices and Experts on Block 1 (Familiar Stimuli) 32 20. Exp. 2— Error Rates for Novices and Experts on Block 1 (Familiar Stimuli) 32 21. Exp. 2— RTs for Novices and Experts on Block 2 (New Stimuli) 35 22. Exp. 2— Error Rates for Novices and Experts on Block 2 (New Stimuli) 35 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. v ii Abstract Variations in surface luminance produced by changes in the direction of illumination and contrast can greatly alter the image o f an object. To what extent is human object recognition invariant to such changes? Subramaniam & Biederman (1997) found a huge cost of changes in the sign of contrast when matching faces but not familiar objects. Object (but not face) matching can generally be accomplished by using nonaccidental properties (NAPs) of edges.. .These properties would be unaffected by changes in illumination or sign of contrast. Exps. 1 and 2 used a matching task with smooth, blobby volumes generated from the harmonics o f a sphere. These objects lack surface discontinuities and, only when highly dissimilar, do they differ in NAPs. Changes in contrast polarity had no effect on stimuli matching for either novices or experts, suggesting that faces are special with respect to sensitivity to sign of contrast. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. Introduction: Reported differences in Object and Face Recognition In a study done by Subramaniam and Biederman (1997) in which subjects completed a same-different sequential matching task on faces and chairs (Figure 1), it was revealed that subjects showed a dramatic decline in their ability to match faces when they differed in contrast polarity. However, no such costs were apparent when Figure 1: Contrast Inverted Chairs and Faces. Examples of face and object stimuli from Subramaniam & Biederman (1997). Note the increased difficulty in recognizing the contrast inverted faces as the same in shape vs. the contrast inverted chairs. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 2 Average Responses - Chairs vs. Faces SI: Normal Com, Rev Normal ContRev S2: N o rm ! Cdnt. Rev Corn. Rev Normal Note the enormous cost to FACE, but not OBJECT, matching, when matching images of different polarity vs same polarity. Figure 2: Results from Subramaniam & Biederman (1997). M ean correct nam ing RTs and error rates averaged over "Same" and "Different" responses. Error bars show the standard errors o f the difference scores between each subject's m ean score and his or her mean score for that condition. matching chairs (Figure 2), even when the chairs and faces were scaled to be equally similar according to a wavelet model of similarity (Lades et al.,1993; Wiskott et al., 1997, Okada, 2002). This model determines stimuli similarity using a graph matching procedure to compare a Gabor wavelet representation o f the two stimuli which are represented as activation values of Gabor kernels in an array o f columns (Figure 3). Each column is termed a “Gabor jet, ” with the receptive fields (R.F.s) of R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. the kernels in each jet centered on a particular position of the face, such as the comer of the eyes. The location and scale of the kernels would ultimately have to be in face rather than world coordinates to allow for translation and scale invariance. This S tored o b ject representation O bject (m em ory) layer Matching algorithm Multidimensional featu re d e te c to r Input (featu re) layer The direction of diffusion Figure 3: Illustration of the Lades et al. (1993) architecture. Each Gabor jet is illustrated as a stack of disks, each disk corresponding to a Gabor filter [at one of 8 orientations and 5 scales) centered at a single lattice position in the visual field which defines its receptive field. The receptive fields of the largest filters were considerably larger than indicated in the Figure in that they were affected by luminance variation approximately two nodes away from the center of their receptive fields. A stored image in the object (memory) layer is matched against a new image by allowing the jets representing the new image to diffuse (change their position from their original position in the lattice) to achieve a pattern of activation over the kernels that is most similar to the pattern for that jet in the memory layer. Dissimilarity is a function of both the degree to which of the activation values for each jet differ from the original and the degree of lattice distortion after diffusion. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 4 type of face representation in terms of activations of hypercolumns of Gabor filters has been argued to explain many phenomena unique to face recognition and not found in object recognition in general (Biederman & Kalocsai, 1997). Why should there exist a difference in the costs of contrast inversion for faces and objects? One possibility is that unlike face matching, object matching (such as the chairs in Subramaniam & Biederman’s experiment) can generally be accomplished by using parts and contours at orientation and depth discontinuities. These features would be unaffected by changes in contrast polarity. Face matching may rely much more on variations in shading used to discern smooth facial surface contours or on face pigmentation information (Hill & Bruce, 1996). Smooth surface structure information would be affected by changes in contrast polarity, which is evident in the reported increase in subjects’ error rates when matching negative contrasted faces but not chairs. Therefore it is possible that the presence of contrast- invariant information in non-face objects accounts for the differences in recognizing these two classes of stimuli. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 5 1. Experiment 1: Novice and Expert Sequential Matching of Blobs Generated bv the Rotational-Variation of the Harmonics of a Sphere. In this experiment, we examined whether object recognition would remain invariant to contrast polarity when the objects were non-face, three-dimensional, smoothly curved, and novel. The discrimination of such objects would appear to require the same kind of information employed when discriminating faces. If subjects showed invariance to contrast polarity for these objects, it would suggest that faces are special with respect to this variable. In addition to examining the effects of contrast polarity on object recognition, we investigated the effects of expertise on these types of object recognition tasks. Gauthier & Tarr (1997) have argued that the sensitivity to sign of contrast (also referred to as “direction” of contrast) in face matching is a consequence of the experience (expertise) we have with face images of positive contrast. In Part 2 of this experiment, we looked at whether or not intensive training on blob matching in positive contrast leads to deficits in matching blobs of different contrast. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 6 1.2 Experiment 1: Methods 1.2.1 Exp. 1 Subjects Eight University of Southern California graduate and undergraduate students (four “Expert” and four “Novice” subjects), age 19-26 years, participated in the study in return for monetary compensation. There were four females and four males, All of these One of these ■ ’ C O One of these Figure 4: Shepard & Cermak’s (1973) procedure for generating a free form. In this specific example, 1) the four fixed components in row 1 are combined with one orientation of 2) the two-lobed component and one orientation of 3)the three-lobed component to generate each stimulus in the smoothly varying space. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 7 all of whom were right-handed. None of the participants had seen the stimuli prior to the experiment. 1.2.2. Exp. 1 Stimuli The stimuli were three-dimensional, asymmetrical, smooth blobby objects created using Matlab on a Macintosh G3 computer. The method for creating the objects was Both of these + One of these + One of these XXX X m Figure 5: Exp. 1- Production of Rotationallv-Varied Blob Stimuli. The second (peanut shape) and third (propeller shape) harmonics of a sphere are added together in different orientations (eight equally-spaced steps between 0-360 degrees) to produce a toroidal stimuli space (the 0 and 360 degree rotation positions of each harmonic produce identical blobs). R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. inspired by Shepard and Cermak’s (1973) method of producing the harmonics of a circle (Figure 4) and modified to create three-dimensional stimuli varying only in metric properties. For our stimuli, a base shape was produced by adding a sphere to a three-dimensional shape with eight equally spaced convex “lobes” (Figure 5). The second and third harmonics of a sphere (three-dimensional shapes with either two or six equally spaced convex “lobes”, respectively) were then added together in different orientations to the base to produce objects with smooth surface protuberances in different orientations. These objects varied gradually from one another in a toroidal space, i.e. the space wraps around itself in both directions (like a doughnut), so that the effects of similarity between all objects in this space could be assessed (Figure 6). Nearby stimuli in this space did not differ in parts or nonaccidental properties due to the fact that they smoothly varied from one another only in their surface curvature along with the eight slight rotations of both of the harmonics. The number of rotation steps was chosen to ensure that the objects could be both more similar and less similar than stimuli used in other studies examining face vs. object recognition (i.e. “normal” faces and “greebles”) The similarity of pairs of objects was scaled R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 9 ■¥ % • / . ^ ......... 1 4 8 12 16 20 24 28 32 ^ m 4 f ^ & , 2 i t K § 4% y A %»* 28 3? ¥ f i ¥ ¥ f t ¥It ¥ * 20 24 28 ^ 3 2 to *** W ’ WWWii'iw I p W & ' m v v&usk¥¥ ¥¥¥ Figure 6: Assessing Similarity Values. Similarity Values as assessed by a Gabor Jet Similarity (GJS) Program of Rotationally-Varied Blobs in a Toroidal Space. Blobs located physically close to one another in this space (the harmonics are rotated by only a few small steps) have a low city-block distance (CBD) and a high GJS Value. In this example, the blobs at (12,4) and (16,1) are two city- block steps away from one another (one rotation of the 2n d harmonic and one of the 3rd , a CBD of zero would be identical) and have a GJS of 95 (100 would be identical). according to a wavelet similarity measure (Lades et al., 1993), as was done for the stimuli used by Subramaniam and Biederman (1997). These similarity values were highly correlated (r=.998) with the physical “city-block distances” (CBD) of the R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. stimuli within the toroidal blob space. The CBD are physical steps in the stimuli space along which the second and third harmonics vary, i.e. two blobs that are one city-block step away from one another would differ by the smallest possible rotation of either the 2n d or 3rd harmonic of a sphere. These values varied from a Gabor Jet Similarity of 65 for the most dissimilar stimuli (a CBD of 8) to 100 for identical stimuli (a CBD of 0). This stimuli space was far larger in size and varied with far smaller changes in appearance than the space comprising typical faces. When this Compare to base: * - t t v t t t H i ik J l k 2B8 k , a t .JBk JBm J t ,|B | J City Block Distance Gabor Jet Similarity (high-->low) Figure 7: City Block Distances and Gabor Jet Similarity Values. Stimuli similarity values in city- block distance (CBD) for Blob stimuli, a variety of “normal” faces varying in appearance and gender, and Greebles. The CBDs are highly correlated (t=.998) with the GJS values, with normal human faces generally being in the range of 2-4 CBDs and Greeble stimuli being in the range of 2-7 CBDs. Note that the blob stimuli space encompasses that of both normal human faces and Greebles, i.e. the most similar blobs (CBD of 1) are more similar than either faces or Greebles, and the most dissimilar (CBS of 8) are more dissimilar. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 11 same method was used to assess the similarity of 14 faces varying in appearance and gender, the Gabor Jet Similarity values fell within the range of 92 for the most similar faces to 79 for the least similar (Figure 7). This reflects the fact that we were examining recognition effects on objects that spanned the full space of similarity possible for typical faces, controlling for the magnitude of stimuli differences or energy. Once each blob was constructed, it was then illuminated with two point- source lights in front (set at infinity to prevent cast shadows), at 45 degrees above the x-axis, and 90 degrees apart on the y-axis. The surface was a uniform matte gray. The rendered images of the objects were then converted into 8-bit grayscale images at 72 dpi using Adobe Photoshop 5.0 and presented on an Apple Macintosh G3 computer at a resolution of 1024 x 768 pixels at a refresh rate of 75 Hz. 1.2.3 Exp. 1 Design The following are the four possible trial types that could vary on each trial. The order of presentation of the trials was balanced across participants so that each object condition appeared equally likely in each trial position (Figure 8). Subjects were not briefed on the frequencies of the types of trials. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 12 A. Same Shape / Same Contrast The first object (SI) always appeared in the same position in the center of the screen. The second object (S2) appeared in one of eight equidistant positions around the center. On 25% of the trials, SI was the same shape as S2 and had the same contrast (positive or negative). All parts of the toroidal stimuli space were presented equally often in each of the shape and contrast conditions. B. Different Shape / Same Contrast On 25% of the trials, the first object (SI) was different from the second object (S2) in shape but were both presented with the same sign of contrast. These trials equally spanned the full range of the similarity values possible in the toroidal stimuli space. C. Same Shape /Different Contrast On 25% of the trials, SI was the same shape as S2 but differed in the sign of contrast. Both SI and S2 were equally likely to be in positive and negative contrast over the full range of similarity values in the stimuli space. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. D. Different Shape/Different Contrast On 25% of the trials, SI was a different shape and contrast than S2. Once again, both SI and S2 were equally likely to be in positive and negative contrast over the full range of similarity values in the stimuli space. Figure 8: Experimental Conditions for Exp. 1 (Rotationally Varied Blobs). Left blobs in each box represent S1, right blobs represent S2. Blobs could either be of the same or different shape and same or different contrast. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 14 1.2.4 Exp. 1 Procedure Part 1: Blob Matching by Novices A sequential-matching paradigm was used, in which participants had to judge whether two sequentially presented objects were the same or different in shape. Each participant ran in 1024 test trials, in which two objects (of the 64 rendered 2nd Stimulus 250 msec 1st Stimulus 400 msec Fixation Cue 500 msec Figure 9: Example “Same” Trial for Experiment 1. Correct response is “Same” because SI and S2 are the same shape, despite the change in sign of contrast. Note the position shift between SI and S2 to make a quick “Same” response difficult when there was no change in stimulus shape or contrast. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 15 blobs) were shown consecutively, with each trial being in one of the four described conditions and the first object (SI) appearing in the center of the screen followed by the second object (S2) appearing in a different position. The eight possible positions for S2 to appear in werepositioned equally around a circle with 45 degrees between each possible position along the circumference. The radius of this circle was calculated to be as large as possible with minimum overlap between the first and second images. Each trial was composed of a fixation cross for 500 ms., the first object (SI) for 400 ms., and the second object (S2) for 250 ms (Figures 9 & 10). The participants were asked to judge as quickly and as accurately as possible whether the two objects in a given trial were the same or different in shape, regardless of any change in contrast or position. They were given 1600 ms. from the time the second object image was presented to respond. If they did not respond by this time, or if their response was incorrect, they heard a loud beep as feedback. If their response was correct and timely, no feedback was given. MacProbe 1.8 software was used to write the program for the experiment, as well as run the experiment and record responses, response times, and subject conditions. The experiment was run using a Power Macintosh G3 computer and a Sony 17 inch R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 16 - w t m M - 31ff llM BC r SCOf l i s ee Figure 10: Example “Different” Trial for Experiment 1. Correct response is “Different” because SI and S2 are different shapes, despite having the same sign of contrast. GDM-500PS Multiscan® Computer Display. The order of presentation of the trials was balanced for each condition across participants so that each object condition appeared equally likely in each trial position. Participants received one rest spaced equally between two trial blocks, and a short practice session at the beginning of the experiment. The practice session was comprised of 32 trials that replicated the test conditions so that the subject became R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 17 familiar with the task. The entire experiment including practice trials, test trials, breaks, and debriefing lasted approximately 60 minutes. Part 2: Blob Matching by Experts Achieving expertise in these types of tasks has been reported to require about 3,240 trials (or 7-10 hrs. of training) on average (Gauthier & Tarr, 1997; Gauthier et al., 1998). To produce expertise in our subjects, they performed eight training sessions of 1,024 trials each consisting only of the positive contrast images, for a total of 8,192 trials and about 8 hrs. of training. After attaining expertise, the “Experts” then completed the testing session with images of both positive and negative contrast, identical to that of the “Novices” (Part 1). Other than the training on the positive contrast images before being tested on images of both contrasts, all conditions were repeated for the Experts. 1.3 Experiment 1: Results & Discussion When computing mean response times, incorrect responses were discarded. In addition, responses whose duration was shorter than 150 ms or longer than 1600 ms were treated as errors. This included about 9% of the trials, which did not differ significantly between the trial types (p=.95) or stimuli similarity (p=.82). To further R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 18 assess whether the “physical stimuli space” (city-block distances, or CBDs) adequately represented the actual perceptual similarity of the stimuli the response times and error rates across the Gabor Jet Similarity values were computed and found to be highly correlated (r=.93 and .94 for response times and error rates, Figures 11 and 12, respectively). Novice vs. Expert Recognition Overall response times and error rates on different trials for both Novices and Experts in the test session as a function of stimuli similarity and contrast direction (same vs. different) are shown in Figures 13 and 14. In the test session, the Experts clearly performed faster and with fewer errors than the Novices. Collapsed over similarity, the mean correct response time for the Novices was 464 ms.; for the Experts, 299 ms. Averaged percent error for the Novices was 30.0%; for the Experts, 17.0%. Same vs. Different Contrast As is evident from inspection of Figs. 13 and 14, there was no effect on response time or error rates of changes in the sign of contrast for either Experts or Novices. Collapsed over similarity, for the same contrast trials, mean correct response time R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 19 500 r * .92 450. 400- £ o § I- 350- c o 1 300- 250 '8 4 ' 82' 80 78 ' 76 74 7 2 ' 70 '68 66 64 62 Low ■ 94 92 90 High 0.7 r = .93 0.6 Chance t l i I 0.4- a 0 .2 - 0.0 90 96 94 92 90 80 86 04 02 80 78 76 74 72 70 68 66 64 62 Qabor Jet Similarity Value (Identical = 100) Gabor Jet Similarity Value (Identical = 100) Figures 11 & 12: GJS, RTs, and Error Rates. Correlations between the Gabor Jet Similarity values for pairs of stimuli and the reaction times and error rates, respectively. Note the high r-values: Similiarity is a good predictor of performance on blob-matching for both reaction times and error rates. for the Novices was 459 ms; for the Experts, 299 ms. Averaged percent error for the Novices was 30.0%; for the Experts, 17.0%. For the different contrast trials, mean correct response time for the Novices was 468 ms; for the Experts, 299 ms. Averaged percent error for the Novices was 32.0%; for the Experts, 17.0%. As similarity decreased on the different shape trials, both groups showed a significant decline in error rates with significant main effect of similarity; for Novices, F(7,14)= 14.4,p=.00, for Experts, F(7,14)=26.9,p=.00. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 20 0.70 -s - Novice Sam e C ontrast -*■ Novice Dlff C ontrast -u- Expert Sam e C ontrast ■*» Expert Dlff C ontrast 0.60 0.50 0.40 0.30 0.20 0.10 0.00 G abor J e t Similarity (high->low) Figure 13: Exp. 1 —RTs for Novices and Experts on Same vs. Different Contrast Trials. Note the superior performance of the Experts and also the lack of separation between the lines representing the same and different contrast trials. 600 Novics Sams Contrast Novice Dlff Contrast Export Sams Contrast Expert Dlff Contrast 550 500 1 450 I P 400 O ! 8 350 c c 300 250 200 Gabor Jet Similarity (hlgh-->low) Figure 14: Exp. 1 —Error Rates for Novices and Experts on Same vs. Different Contrast Trials. Again, note the superior performance of the Experts, and the lack of a contrast effect, as well as a main effect of similarity (both groups perform better on low similarity stimuli.) R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 21 Same vs. Different Shape When examining the trials in which SI and S2 had the same or different shape, regardless of their sign of contrast, neither the Experts nor the Novices showed a significant difference in their response times (Figure 15). The mean response times for the Experts on “Same Shape” trials was 268 ms and on “Different Shape” trials was 293 ms., and for the Novices on “Same Shape” trials was 428 ms and on “Different Shape” trials was 490 ms. However, there was a significant effect of the Shape condition on both groups’ error rates (Figure 16); for Novices, F(l,3)=6.81, p=.02, for Experts, F(l,3)=12.47, p=.04. The mean error rates for the Experts on “Same Shape” trials was 8.2% and on “Different Shape” trials was 19.7%, and for the Novices on “Same Shape” trials was 12.5% and on “Different Shape” trials was 28.4%. A post-hoc Scheffe analysis of the interaction between the Shape and Contrast conditions indicated that the difference between “Same” and “Different” contrast trials was largest when SI and S2 had the same shape (for Novices response times and error rates respectively, p=.01 and p=.025; for Experts, p=.74 and p=.01), and that this interaction was more significant for the Novices than for the Expert subjects. This could be a criterion effect of subjects being more likely R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 2 2 600 Novice Same Contrast Novice D tfT Contrast Expert Same Contrast Expert D tfT Contrast 500 - " 3 E 450 i * 400 o s 350 a 300 250 200 Different Shape Same Shape Stimuli Shape (SI &S2) Figure 15: Exp. 1 —RTs for Novices and Experts on Same vs. Different Shape Trials across contrast conditions. Neither Novices nor Experts show an effect for the Shape or Contrast conditions. a c c 0.70 Novice Same Contrast Novice Dlff Contrast Expert Same Contrast Expert Diff Contrast 0.40 0.10 0.00 Same Shape Different Shape Stimuli Shape (SI &S2) Figure 16: Exp. 1 —Error Rates for Novices and Experts on Same vs. Different Shape Trials across contrast conditions. Both Novices and Experts show an effect for the Shape, but not Contrast, conditions. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 23 to respond “Same” than “Different”, an issue which is addressed by the use of a new task in Experiment 2. 2. Experiment 2: Novice and Expert Forced-choice Match-to-sample of Blobs Generated bv the Amplitude-Variation of the Harmonics of a Sphere. In this second experiment, we assessed whether object matching would remain invariant to contrast polarity if training was confined to an extremely restricted part of the stimulus space in which only the amplitudes of the second and third harmonics varied, but not their orientation. This manipulation was done to simulate the variation among face parts, e.g., noses, chins, etc., that can vary in size much more than in relative position. Thus noses are typically centered under the eyes and the mouth is centered under the nose, with little variation. The rotation of the second and third harmonics, particularly at lower similarity values, produced different protuberances and variation in the relative position of these structures. The sequential matching task used in Exp. 1 allowed an effect of criterion in which very high similarity values on different shape trials were responded to at or below chance, reducing the chance of obtaining an interpretable effect of contrast R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 24 polarity at these similarity values. Consequently, in this experiment we employed a match-to-sample task that avoided this criterion problem. 2.2 Experiment 2: Methods 2.2.1 Exp. 2 Subjects Eight University of Southern California undergraduate students, ages 17-23 years, participated in the study in return for monetary compensation (four “Experts” and four “Novices”). All subjects were right-handed, including six females and two males. None of the participants had seen the stimuli prior to the experiment. 2.2.2 Exp. 2 Stimuli From the original toroidal stimuli space used in Exp. 1, four equidistant blobs were chosen, and the amplitudes of the harmonics for each was metrically varied (Figure 17) while their orientations were maintained, to produce four new stimuli spaces (Figure 18). In each of these four new spaces, only the size of the harmonics varied (degree of surface curvature), so that all of the stimuli within each space had the same configuration of convexities and concavities. In all other ways, these stimuli were similar to the previous stimuli in Experiment 1, i.e. they were smooth, blobby, asymmetric, non- R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 25 face, novel volumes, and were produced using the same programs under the same conditions. Both of these + One of these + One of these * Figure 17: Exp. 2-Production of Amplitude-Varied Blob Stimuli. The second (peanut shape) and third (propeller shape) harmonics of a sphere are added together in the same orientation but different sizes (8 different size steps for each harmonic, similar to the production of the blob stimuli in Exp. 1). This produced a stimuli space containing 64 different blobs with the same harmonic configuration but with different harmonic amplitudes. This was done for four different harmonic configuration to produce four different amplitude-varied stimuli spaces. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. Figure 18: Exp. 2 — Stimuli (amplitude-varied blobs). Four Blob spaces generated by the variation of the amplitudes of the 2n d harmonic (along the x-axis) and the 3rd harmonic (along the y-axis) for four different harmonic configurations. 2.2.3 Exp. 2 Design There were four possible conditions that could vary on each trial. The order of presentation of the trials was balanced for each condition across participants so that each object condition appeared equally likely in each trial position (see Figure 19). R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 27 A. Same Contrast (allpositive) On 25% of the trials, all objects were of positive contrast. B. Same Contrast (all negative) On 25% of the trials, all objects were of negative contrast. C. Different Contrast (target blob is o f positive contrast) On 25% of the trials, the target object was of positive contrast and the two sample objects were of negative contrast. D. Different Contrast (target blob is o f negative contrast) On 25% of the trials, the target object was of negative contrast and the two sample objects were of positive contrast. Due to an error in balancing, half of the testing trials (positive and negative contrast) were run with the two sample (lower) stimuli in different contrast, allowing a bias effect simply based on matching the direction contrast. These trials were ignored in the analysis. Such trials were also not included in the later match-to- sample experiments. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 2 8 mAINING: Same Contrast BLOCK!: Draining Blobs BLOCK 2: New Blobs 3 I s |J w c Ika I m A A I s i I s I 2 ,5 > DC I Figure 19: Experimental Conditions in Exp. 2 (Amplitude-Varied Blobs). In each tile, the top blob is the “Target” blob shape, and subjects are to decide which of the bottom “Sample” blobs matches the “Target” in shape. Experts performed 8,192 training trials in just positive contrast, followed by a testing session of 1) 512 trials of blobs with the same harmonic configuration as the training trials in both positive and negative contrast (Block 1), and then 2) 512 trials of blobs with a new harmonic configuration in both positive and negative contrast (Block 2). Novices completed the same type of testing session as the Experts in both positive and negative contrast, but without any prior training of blobs of just positive contrast. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 29 2.2.4 Exp. 2 Procedure A forced-choice match-to-sample task was used, in which three objects were presented simultaneously and participants had to judge which of the bottom two objects (the “samples”) matched the top object (the “target” ) in shape. This task would not be affected by memory, i.e., it minimizes memory demands because the stimuli are presented on the screen simultaneously. It also has no possible criterion effects or response biases forjudging “same” vs. “different”. Each trial was composed of the three objects being shown simultaneously for 1 s. The participants were asked to judge as quickly and as accurately as possible which of the two bottom “sample” objects matched the top “target” object in shape, regardless of any change in contrast. They were given 3 s. from the time the objects were presented to respond. If they did not respond by this time, or if their response was incorrect, they heard a low-pitched ’’ beep” as feedback and the trial was recorded as an error. If their response was correct and timely, they heard a high- pitched “beep”. Each of the subjects ran in 1024 trials during the “testing” session, with all stimuli appearing equally often in each of the testing conditions. Prior to performing the “testing” session, the Experts performed eight 1,024 trial “training” sessions in R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 30 just positive contrast, but in all other ways duplicating the experimental conditions of the “testing” session. The Novices received no such prior training on positive contrast stimuli. The experiment was run on the same platform and under the same conditions as Experiment 1. The order of presentation of the trials was balanced for each condition across participants so that each object condition appeared equally likely in each trial position. Participants received two rests spaced equally between three trial blocks, and a short practice session at the beginning of the experiment. The practice session was comprised of 32 trials that replicated the test conditions so that subjects would be familiar with the task. The entire experiment including practice trials, test trials, breaks, and debriefing lasted approximately 60 minutes. The only other deviation in procedure from Experiment 1 was on the testing session (blobs of both positive and negative contrast) for both Novices and Experts. On the first half of this session (Block 1:512 trials), the blobs were amplitude variations of the 2n d and 3rd harmonics of the same blob defined by a particular orientation of the harmonics (see. Figure 18). The same space as used during the training sessions was presented to Experts (“familiar” space). During the second half (Block 2:512 trials), the blobs were amplitude variations from a different shaped R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 31 blob (the different shape produced by different orientations of the second and third harmonics). These were presented to both Novice and Expert subjects. The presence of four different stimulus spaces composed of objects which varied in their metric surface curvature in the same way made it possible to test for a transfer of expertise to a blobs of a different configuration (shape). Both Expert and Novice performance could be compared on new and old stimulus spaces. 2.3 Experiment 2: Results & Discussion When computing mean response times, incorrect responses were discarded. In addition, responses whose duration was shorter than 150 ms. or longer than 3 s were discarded as mistrials. This included less than 1% of the trials, which did not differ significantly between the trial types or stimuli similarity. Initial analysis involved the computation of mean response times and error rates for Same contrast vs. Different contrast conditions averaged over blob similarity values. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 32 1300 Novice Same Contrast Novice Dlff Conlrast Expert Same Contrast Expert Dlff Contrast 1250 1200 1150 1 1100 ~ 1050 * > 1 1000 i 950 900 800 Gabor Jet Similarity <high->low> Figure 20: Exp. 2—RTs for Novices and Experts on Block 1 (Familiar Stimuli). Note the superior Expert performance and the lack of separation between the lines representing Same vs. Different contrast trials. 0.70 Novice Same Contrast Novice Dill Conlrast Expert Same Contrast Expert DU! Contrast 0.60- 0.1 0- 0. 00 Gabor Jet Similarity (hlgh->lo«0 Figure 21: Exp. 2— Error Rates for Novices and Experts on Block 1 (Familiar Stimuli). Note again the superior Expert performance, a lack of an effect of contrast direction, and a significant effect of stimuli similarity (better performance on dissimilar objects). R eproduced with perm ission o f the copyright owner. Further reproduction prohibited without perm ission. 33 Block 1 (Familiar Space): Collapsed over similarity, for the same contrast trials, mean correct response time for the Novices was 1080 ms; for the Experts 871 ms. Percent error for the Novices averaged 27.0%; for the Experts, 17.0%. For the different contrast trials, mean correct response time for the Novices was 1079 ms; for the Experts, 877 ms. Percent error for the Novices averaged 28.0%; for the Experts, 17.0%. The Experts performed much better than the novices as reflected in both the response times and error rates, indicating that expertise was definitely attained in our subjects (Figures 20 and 21 for Response Times and Error Rates, respectively). There was no effect on response times or error rates of changes in the sign of contrast for either Experts or Novices, F(l,6)=.00,p=.97 andF(l,6)=.00,p=.95, respectively. There was a significant effect of the level of similarity of the stimuli to be matched for Experts and Novices; for Experts, response times: F(7,21)=13.1, p=.00, error rates: F(7,21)=44.5, p=.00; for Novices, error rates: F(7,21)=18.25, p=.00. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 34 Block 2 (New Space): Collapsed over level of similarity for the same contrast trials, mean correct response time for the Novices was 1007 ms; for the Experts, 873 ms. Percent error for the Novices averaged 28.0%; for the Experts, 21.0%. For the different contrast trials, mean correct response time for the Novices was 1012 ms; for the Experts, 889 ms. Percent error for the Novices averaged 27.0%; for the Experts, 21.0%. The Experts showed no cost at all for matching blobs from the new blob space, indicating that they transferred their expertise to a new stimuli space, F(l,3)= .00, p=.95. The Novices performed slightly better during Block 2 (new stimuli), an effect likely attributable to the training on Block 1. Lastly, and most important, there was no effect on response times or error rates of changes in the sign of contrast for either Experts or Novices, (Figures 22 and 23 for Response Times and Error Rates, respectively). R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 35 1300 Novice Same Contrast Novice Dlff Contrast Expert Same Contrast Expert Dlff Contrast 1250 1200 1 1100 1050 | 1000 a 050 900 8S0 800 Gabor Jet Similarity <high->low) Figure 22: Exp. 2— RTs for Novices and Experts on Block 2 (New Stimuli). Note that the Experts performed at the same level as in Block 1 and there was no effect of contrast inversion. 0.70 Novice Same Contrast Novice Dlff Contrast Expert Same Contrast Expert Dlff Contrast 0.60 0.10 0. 00 Gabor Jet Similarity <high->low) Figure 23: Exp. 2— Error Rates for Novices and Experts on Block 2 (New Stimuli). Again note the Experts’ maintained performance, the lack of an effect of contrast, and a significant effect of similarity. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 36 General Discussion The goal of these experiments was to assess the effects of differences in contrast polarity on the matching of non-face objects which required the use of information presumed to mediate the matching of faces. In Experiment 1, we found an insensitivity to contrast direction when matching novel objects. In this case, the only information available to complete the task were the metric changes in the smooth, surface curvature of the objects, similar to the information available when making discriminations between faces, but not when distinguishing most subordinate level objects. This is also consistent with much research implicating that the representations used to make each of these types of discriminations might be different (Biederman & Kalocsai, 1997; Tanaka & Farah, 1993). The invariance to contrast was found in both Novice and Expert subjects, ruling out the possibility that sensitivity to contrast could be due to holistic processing based on vast training with a particular set of stimuli (“Expertise”) as suggested by Gautier & Tarr (1998) and Gauthier et al. (1998). Our subjects did show sensitivity to the level of similarity of the stimuli indicating that they were indeed making fine, metric discriminations between the blobs’ structure, based on the similarity and spatial relations of the R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 37 “features” (protuberances), presumably in the same way faces are discriminated (Rhodes, 1988). Many researchers arguing against differential processing systems for face and object recognition have claimed that differences found between the processing of these two types of stimuli (such as the ability to process upside-down or contrast- inverted objects, but not faces,) are actually due to expertise that humans develop in differentiating members of the subordinate category of human faces (for this argument with inverted faces and objects, see Carey and Diamond, 1977). However, it has been demonstrated that faces that have been manipulated in these ways (orientation and contrast inversion) are processed through part decomposition, like non-face objects, and not holistically, as is the case with normal, upright faces (for rearranged faces, Tanaka & Farah, 1993; for inverted faces, Yin, 1970; for contrast inverted faces, Subramaniam and Biederman, 1998 for face recognition in noise, McKone, Martini, & Nakayama, 2001). We had trained our Expert subjects far more extensively than other researchers had reported was necessary to gain “expertise” in novel object discrimination, and the Experts were performing significantly better than the Novices: evidence that they had gained expertise. In Exp. 2, we trained our expert subjects for the same amount R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 38 of trials under similar conditions as in Exp. 1, but in this experiment, as a further test of their expertise, they were tested on novel shapes. They demonstrated perfect translation of their expertise to this new stimuli space, as well as lower response times and error rates as compared to the novices, indicating that they had truly gained expertise and still failed to show sensitivity to contrast inversion. This evidence for contrast invariance is more unequivocal than previous research attempting to show sensitivity to contrast based on expertise. Many of those studies relied on stimuli which resembled faces (like “Greebles” which have features similar to facial features) and tasks not testing pure discrimination (like “naming” tasks based on memory and possibly dependent on the stimuli’s sign of contrast as a cue to the object’s identity), (Gauthier et al., 1998). By using a match-to-sample task we were able to test object matching without criterion effects or memory. We assessed the effects of expertise, not only through performance improvement but also through a transfer to a new stimuli set. This research also developed novel, non-face stimuli whose discrimination can only be accomplished by using small, metric differences in smooth surface curvature. Overall, this research used much more stringent methods for assessing any possible effects of contrast inversion on the R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 39 matching of non-face objects, but failed to do so, lending much credence to the assertion that faces are a special category of objects. In addition to behavioral studies such as this one reporting differences in face and object processing, there are several other branches of research with notable evidence for differential processing of these two categories of stimuli. Some of the earliest and strongest evidence of the existence of two different but related systems for processing objects and faces is the presence of two different types of visual agnosias based either on the inability to process either faces or objects largely independently of one another. Prosopagnosia, the inability to recognize familiar people by their faces, has been most recently demonstrated to be present following lesions to the inferior occiptotemporal region (Damasio, Damasio, & Van Hoesen, 1982; Meadows, 1974). It can co-occur with additional deficits of recognition of other non-face objects (indicative of perhaps a more general deficit in recognition), but it can also be present on its own (Dixon, Bub, & Arguin, 1998; Farah, Wilson, Drain, & Tanaka, 1995). Object agnosia is the inability to recognize or categorize non-face objects. This can involve all or just specific categories of objects, and can co-occur with prosopagnosia, but doesn’t have to. Since these agnosias appear to be relatively independent of one another, and patients displaying these agnosias can R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 40 have otherwise highly functioning visual recognition, these agnosias are likely due to a true categorical deficit and not a deficit of general visual processing (Caramazza & Shelton, 1998). In addition, prosopagnosics do not show the same costs for matching inverted faces that “normal” subjects do, in fact, they show an advantage for matching such stimuli as compared to “normal” subjects (Farah, Wilson, Drain, & Tanaka, 1995). This indicates that they have damage to a specialized face recognition system that cannot be utilized as it is by the “normal” subjects, even though this is ultimately to their advantage when matching inverted faces. Another source of evidence for different processing of objects and faces comes from reports of differential activity in various parts of the brain when processing these two classes of stimuli. As one examines the response of neurons in the inferior temporal cortex to patterns and objects (specifically, moving forwards from the area of TEO [posterior inferior temporal cortex] towards TE [anterior inferior temporal cortex]), it is evident that some their responses selectively increase with the increasing complexity of features of particular classes of patterns and objects (Tanaka, 1993; Kobatake & Tanaka, 1994). The cells that respond selectively to faces alone are referred to as face-selective cells, and have been reported in the superior temporal sulcus (STPa) (Yamane et al 1988, Perrett et al R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 41 1982), TE (Tanaka et al, 1991), and the amygdala (Nakamura et al 1992). These cells respond not just too whole faces, but also face parts, and particular views of faces (Perret et al., 1987; 1990; Desimone, 1991; Gross, 1992; Logothetis & Scheinberg, 1996). Work done on human subjects recording event-related potentials (ERPs) from the surface of occipitotemporal extrastriate cortex evoked by visual stimuli has shown that this area was activated by faces, but not by other types of visual stimuli (Ungerleider & Mishkin, 1982), which is strong evidence for a specialized face processing area. Lastly, neuroimaging studies using positron emission tomography (PET) have also shown differential activity for faces in the anterior and posterior fusiform gyrus (Haxby et al., 1993). The present study thus joins these other investigations in documenting that faces, as a class of stimuli, are indeed “special”. We have shown that both Novice and Expert subjects’ performance on matching non-face objects with face-like surface structure is invariant to contrast reversal. If face discrimination can only be accomplished using this same type of surface curvature information, this would suggest that faces are special. However, in addition to surface curvature, pigmentation information, such as high contrast patches including the eyebrows and the shadows of the nostrils, may also be used in face recognition (Bruce & Langton, R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 1994; Troje & Bulthoff, 1996). Future research should examine the effects of contrast reversal on the discrimination of objects who only differ in either surface structure (also referred to as “shape” by other investigators) or pigmentation (also referred to as “texture”). In addition, the dimensionality of information used to discriminate faces may be higher than to discriminate objects (Vetter & Troje, 1997). This would indicate that future research examining face vs. object recognition should utilize stimuli whose discrimination relies on a similar dimensionality to that of faces. If the discrimination of such non-face objects with face-like shape and pigmentation information is also invariant to contrast reversal, this would be further evidence that faces are a “special” type of stimuli with regard to this variable. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 43 References Biederman, I. & Kalocsai, P. (1997). Neurocomputational bases of object and face recognition. Philosophical Transactions of the Royal Society of London B. 352. 1203-19. Biederman, I. & Subramaniam, S. (1997). Predicting the shape similarity of objects without distinguishing viewpoint invariant properties (VIPs) or parts. Investigative Ophthalmology & Visual Science. 38. 998. Bruce, V., & Langton, S. (1994). The use of pigmentation and shading information in recognizing the sex and identities of faces. Perception. 23. 803-822. Caramazza, A. & Shelton, J. (1998). Domain-specific knowledge systems in the brain: the animate-inanimate distinction. Journal of cognitive Neuroscience. 10:1. 1- 34. Carey, S. & Diamond, R. (1977). From piecemeal to configuration processing of faces. Science. 95. 312-314. Cave, C., & Kosslyn, S. (1993). The role of parts and spatial relations in object identification. Perception. 22. 229-248. Damasio, A., Damasio, H., & van Hoesen, G. (1982). Prosopagnosia: anatomic basis and behavioral mechanisms. Neurology. 32. 331-341. Desimone, R (1991). Face-selective cells in the temporal cortex of monkeys. .Tournal of Cognitive Neuroscience. 3. 1-8. Dill, M. & Edelman, S. (2001). Imperfect invariance to object translation in the discrimination of complex shapes. Perception. 30. 707-724. Dill, M. & Fahle, M. (1997). The role of visual field position in pattem- discrimination learning. Proceedings of the Roval Society of London. B. 264.1031- 1036. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 4 4 Dixon, M., Bub, N., & Arguin, M. (1998). Semantic and visual determinants of face recognition in a prosopagnosic patient. Journal of Cognitive Neuroscience. 10:3. 362-76. Farah, M., Wilson, K., Drain, H., & Tanaka, J. (1995). The inverted face inversion effect in prosopagnosia: Evidence for mandatory, face-specific perceptual mechanisms. Vision Research. 14.2089-93. Foster, D. H. & Kahn, J. I. (1985). Internal representations and operations in the visual comparison of transformed patterns: effects of pattern point-inversion, positional symmetry, and separation. Biological Cybernetics. 51. 305-12. Gauthier, I. & Tarr, M. (1998). Becoming a “greeble” expert: exploring mechanisms for face recognition. Vision Research. 37. 12, 1673. Gauthier, I., Williams, P., Tarr, M., & Tanaka, J. (1998). Training “greeble” experts: a framework for studying expert object recognition processes. Vision Research. 38.2401. Gross, C. (1992). Representation of visual stimuli in inferior temporal cortex. Philosophical Transactions of the Roval Society of London B. 335. 3-10. Haxby, J., Grady, C., Horwitz, B., Salerno, J., Ungerleider, L., Mishkin, M., and Schapiro, M (1993). Dissociation of object and spatial visual processing pathways in human extrastriate cortex. In B. Gulyas, D. Ottoson, & P.E. Roland (Eds.), Functional organization of the human visual cortex. Oxford: Pergamon Press, pp.329-340. Hill, H., and Bruce, V.(1996). Effects of lighting and the perception of facial surfaces. Journal of Experimental Psychology: Human Perception and Performance. 22, 986-1004. Kobatake, E., & Tanaka, K. (1994). Neuronal selectivities to complex object features in the ventral visual pathway of the macaque cerebral cortex. Journal of Neurophvsiologv. 71. 856-67. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 45 Lades, M., Vortbriiggen, J. C., Buhmann, J., Lange, J., von der Malsburg, C., Wiirtz, R. P. & Konen, W. (1993). Distortion invariant object recognition in the dynamic link architecture. IEEE Transactions on Computers. 42. 300-311. Logothetis, N., Sheinberg D. (1996). Visual object recognition. Annual Review of Neuroscience. 19. 577-621. McKone, E., Martini, P., Nakayama, K. (2001). Categorical perception of face identity in noise isolates configural processing. Journal of Experimental Psychology: Human Perception & Performance. 27131., 573-599. Meadows, J. (1974). The anatomical basis of prosopagnosia. Journal of Neurology. Neurosurgery, and Psychiatry. 37.489-501. Nakamura, K., Mikami, A., Kubota, K. (1992). Activity of single neurons in the monkey amygdala during performance of a visual discrimination task. Journal of Neurophvsiologv. 67. 1447-63. Nederhouser, M., Mangini, M., Biederman, I., & Okada, K. (2002, November). Matching face-like objects is invariant to differences in direction of contrast. Talk presented at the OP AM: 10th Annual Workshop on Object Perception and Memory, Kansas, MO. Nederhouser, M., Mangini, M., & Biederman, I. (2002, May). The Matching of Smooth. Blobbv Objects— but not Faces— Is Invariant to Differences in Contrast Polarity for both Naive and Expert Subjects. Talk presented at the VSS: Vision Sciences Society conference, Sarasota, FL. Nederhouser, M., Mangini, M., & Biederman, I. (2001, November). Object— but not face— matching is invariant to differences in contrast polarity for both naive and expert subjects. Poster presented at the OP AM: 9th Annual Workshop on Object Perception and Memory, Orlando, FL. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 4 6 Nederhouser, M., Mangini, M., Biederman, I. Subramaniam, S., and Vogels, R. (2001, May) Is Object Recognition Invariant to Direction of Illumination and Direction of Contrast? Poster presented at the 8th Joint Symposium on Neural Computation, La Jolla, CA. Nederhouser, M., Mangini, M., Biederman, I., Subramaniam, S., & Vogels, R. (2001, May) A translation between S1 and S2 eliminates costs of changes in the direction of illumination in object matching. Poster presented at the VSS: Vision Sciences Society conference, Sarasota, FL Okada, Kazunori. Analysis, synthesis and recognition of human faces with pose variations. Dissertation Abstracts International: Section B: The Sciences & Engineering Vol 62(9-B), Apr 2002, 4084. Perrett, D., Rolls, E., Caan, W. (1982). Visual neurons responsive to faces in the monkey temporal cortex. Experimental Brain Research. 47. 329-42. Rhodes, G. (1988). Looking at faces; First-order and second-order features as determinants of facial appearance. Perception. 17. 43-63. Shepard, R. N., & Cermak, G.W. (1973). Perceptual-cognitive explorations of a toroidal set of free-form stimuli. Cognitive Psychology. 4. 351-377. Subramaniam, S. & Biederman, I. (1997). Does contrast reversal affect object identification. Investigative Ophthalmology & Visual Science. 38,998. Tanaka, J., & Farah, M. (1993). Parts and wholes in face recognition. Quarterly Journal of Experimental Psychology. 46. 225-245. Tanaka, K., Saito, H., Fukada, Y., & Moriya, M. (1991). Coding visual images of objects in the inferotemporal cortex of the macaque monkey. Journal of Neurophvsiologv. 66. 170-89. Tarr, M. J., Kersten, D., & Biilthoff, H. H. (1998). Why the visual recognition system might encode the effects of illumination. Vision Research. 38. 2259-2275. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 47 Troje, N. & Biilthoff, H. (1996). Face recognition under varying poses: The role of texture and shape. Vision Research. 36 (12). 1761-1771. Ungerleider, L. & Mishkin, M. (1982). Two cortical visual systems. In: Analysis of Visual Behavior. Ingle, D., Goodale, M., Mansfield, R., (Eds), Cambridge, MA: MIT Press, pp. 549-586. Vetter, T. & Toije, N. (1997). Separation of texture and shape in images of faces for image coding and synthesis. Journal of the Optical Society of America. 14 (9 \ 2125. Wiskott L., Fellous, J.-M., Kruger, N., & von der Malsburg, C. (1997). Face recognition by elastic bunch graph matching. IEEE Trans. Pattern Analysis Machine Intelligence. 19. 775. Yamane, S, Kaji, S., & Kawano K. (1988). What facial features activate face neurons in the inferotemporal cortex of the monkey? Experimental Brain Research. 73* 209-14. Yin, R. (1970). Face Recognition by brain injured patients: A dissociable ability? Neurops vcholo gia. 8. 395. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Face classification

PDF

Behavioral and neural investigations of perceptual affect

PDF

Asymmetries in the bidirectional associative strengths between events in cue competition for causes and effects

PDF

An examination of affective modulation, psychopathy, and negative schizotypy in college and community samples

PDF

Adaptation to sine-wave gratings selectively reduces the sensory gain of the adapted stimuli

PDF

Insights into the nature of phonological and surface dyslexia: Evidence from a novel word learning task

PDF

Articulated thoughts about intentions to commit anti-gay hate crimes

PDF

Cognitive functioning and dementia following cancer: A Swedish twin study

PDF

Articulated thoughts regarding cognitions toward older adults

PDF

Hedonic aspects of conditioned taste aversion in rats and humans

PDF

Alcohol expectancies and consumption: Age and sex differences

PDF

Effect of affective reactions by an outgroup on preferences for crossed categorization discussion partners

PDF

Head injury and dementia: A co-twin control study of Swedish twins

PDF

Emergent literacy differences in Latino and African American children: Culture or poverty?

PDF

Hyperactive symptoms, cognitive functioning, and drinking habits

PDF

Adolescents' social attitudes: Genes and culture?

PDF

A model for figure -ground segmentation by self -organized cue integration

PDF

Auditory word identification in dyslexic and normally achieving readers

PDF

Analysis, synthesis and recognition of human faces with pose variations

PDF

Cross-cultural analysis in anxiety, positive mood states and performance of male cricket players

Asset Metadata

Creator Nederhouser, Marissa (author)

Core Title Invariance to changes in contrast polarity in object and face recognition

School Graduate School

Degree Master of Arts

Degree Program Psychology

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag OAI-PMH Harvest,psychology, cognitive

Language English

Contributor Digitized by ProQuest (provenance)

Advisor Biederman, Irving (committee chair), Baker, Laura (committee member), Lu, Zhong-Lin (committee member), Tjan, Bosco (committee member)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-305679

Unique identifier UC11341536

Identifier 1416569.pdf (filename),usctheses-c16-305679 (legacy record id)

Legacy Identifier 1416569.pdf

Dmrecord 305679

Document Type Thesis

Rights Nederhouser, Marissa

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA