Tilt aftereffect for texture edges is larger than in matchedsubjective edges, but both are strong adaptors ofluminance edges

School of Psychology, University of Wales,Bangor, United KingdomSarah J. Hawley

Department of Optometry, University of Bradford,Bradford, United KingdomDavid R. Keeble

The tilt aftereffect (TAE) has been used previously to probe whether contours defined by different attributes are subservedby the same or by different underlying mechanisms. Here, we compare two types of contours between texture surfaces,one with texture orientation contrast across the edge (orientation contrast contour; OC) and one without, commonly re-ferred to as a subjective contour (SC). Both contour types produced curves of TAE versus adapting angle displaying typicalpositive and negative peaks at ¨15 and 70 deg, respectively. The curves are well fit by difference of Gaussian (DoG) func-tions, with one Gaussian accounting for the contour adaptation effect and the other accounting for the texture orientationadaptation effect. Adaptation to OC elicited larger TAEs than did adaptation to SC, suggesting that they more effectivelyactivate orientation-selective neurons in V1/V2 during prolonged viewing. Surprisingly, both contour types adapted a lu-minance contour (LC) as strongly as did an LC itself, suggesting that the second-order orientation cue contained in thetexture edge activates the same set of orientation-selective neurons as does an LC. These findings have implications forthe mechanisms by which the orientations of texture edges and SCs are encoded.

Keywords: tilt aftereffect, texture edges, subjective contours

Introduction

Texture surfaces in natural visual scenes are oftenbounded by contours composed of aligned line endingswhere surfaces abut or overlap. Textures separated by aboundary contour in this way may differ in properties suchas their orientation or frequency spectra. Such boundariesare often referred to as Btexture edges.[ However, anothertype of texture edge occurs when the two abutting texturesare identical. In this case, the boundary between them isrevealed by aligned line endings only, as the surfaces oneither side are identical. This type of contour is moreusually described as a subjective contour (SC). The com-parison of SCs of this type with the case where adjacenttextures surfaces differ in orientation is rarely made as thetwo types of contour, Bat first glance,[ appear to be so dif-ferent. The detection of orientation contrast, for example,across a border, is considered to be Bpreattentive,[ that is,strongly perceived without conscious effort (cf. Treisman,1985). In comparison, the appearance of SCs is usuallyconsidered to be more phenomenal, perhaps dependent onobserver interpretation, and not unambiguously preatten-tive (for a review, see Pritchard & Warm, 1983). Hence,contours defined by texture orientation contrast would ap-pear to be Bstronger[ visual stimuli in some way.Nevertheless, despite differences in their perceptual

qualities, the two types of contour have some important

physical similarities. In both cases, there are no local first-order cues as to the orientation of the contour, such as aluminance or color discontinuity. Integration of local in-formation is necessary for the orientation percept of evena luminance contour (LC), as the initial input of visual in-formation is in the form of discrete localized points ofinformation, which cannot unambiguously signal globalproperties (Nothdurft & Li, 1985). However, in the caseof an LC, its orientation could be conceived to be the re-sult of Blinking[ of first-order information already rep-resented at a local level. For both of the texture surfaceboundaries described, the orientation attribute of theboundary is clearly a result of processing subsequentto the initial input. In many ways, both types of textureboundary, both with and without orientation contrast,could be described as Bsubjective[ (cf. Bergen, 1991).However, for the remainder of this paper, texture edgeswith orientation contrast will be described as Borientationcontrast contours[ (OCs) and those without will be de-scribed as subjective contours (SCs).Despite the well-documented phenomenon of effortless

texture segmentation due to orientation contrast (Beck, 1966;Julesz, 1981; Julesz & Bergen, 1983; Landy & Bergen,1991; Nothdurft, 1985), it is not clear that the strong vis-ibility of a region of different orientation necessarily im-plies that orientation contrast across a texture edge elicitsa stronger border percept than without, that is, whereonly an SC remains. SCs and texture orientation-defined

Journal of Vision (2006) 6, 37–52 http://journalofvision.org/6/1/4/ 37

doi: 10 .1167/6 .1 .4 Received March 23, 2005; published January 18, 2006 ISSN 1534-7362 * ARVO

contours show similar orientation discrimination thresh-olds (Regan, 1995; Vogels & Orban, 1987; Westheimer& Li, 1996, 1997), and the orientation of both types ofcontour may first be encoded in the response of V2 neu-rons (Baumann, van der Zwan, & Peterhans, 1997; Leventhal,Wang, Schmolesky, & Zhou, 1998; von der Heydt &Peterhans, 1989; von der Heydt, Peterhans, & Baumgartner,1984). While differences in perceptual qualities suggestthat the two types of second-order contour described maydiffer in their processing, the end result of the process-ing in terms of the site of contour orientation encodingmay not be different. Hence, we pose the question, doesthe addition of orientation contrast across the textureedge alter the encoding of contour orientation in the finalpercept?We address this question using the tilt aftereffect (TAE),

a misperception in orientation following prior viewing ofan oriented stimulus. The TAE and related tilt illusion (TI)have both been used extensively to probe whether contoursdefined by different attributes are subserved by the same orby different underlying mechanisms (Berkley, Debruyn, &Orban, 1994; Bockisch, 1999; Clifford, Pearson, Forte, &Spehar, 2003; Georgeson & Schofield, 2002; Paradiso,Shimojo, & Nakayama, 1989; Poom, 2000; van der Zwan&Wenderoth, 1994). The basic premise of such experimentsis that changes in perception of a test stimulus causedby a previously or simultaneously presented Badaptation[stimulus are attributable to a common stage of orienta-tion representation of the two stimuli. In the case of theTAE, prolonged viewing of an Badapting[ line or grat-ing causes a shift in the perceived orientation of asubsequently viewed Btest[ line/grating. A maximal re-pulsive effect is seen for adapt and test stimuli differ-ing in orientation by 10Y20 deg, using either gratings(Poom, 2000; van der Zwan & Wenderoth, 1995) or lines(Blakemore, Carpenter, & Georgeson, 1970; Carpenter &Blakemore, 1973; Paradiso et al., 1989; Skottun, Johnsen,& Magnussen, 1981). A less prolonged misperceptionof test angle can also be seen even with very shortBadaptation[ presentations (e.g., Harris & Calvert, 1989).This repulsive misperception has been termed theBdirect[ effect and is generally accepted to be the resultof altered patterns of activity in orientation-selectiveneurons in V1/V2, most likely due to inhibitory interac-tions (Blakemore et al., 1970; Carpenter & Blakemore,1973; Magnussen & Kurtenbach, 1980a, 1980b; Morrone,Burr, & Maffei, 1982; Wenderoth & Johnstone, 1987).Hence, the existence and magnitude of the direct effectbetween contour types can be considered a measure oftheir shared sites of encoding in at this early level.An attractive misperception, whereby the test orien-

tation appears shifted towards the adapting orientationsignal, is sometimes also seen. This effect peaks at anadapt-test orientation difference of approximately 70 deg.The mechanism of this Bindirect[ effect is more disputed.It may also be due to inhibitory interactions, but in this

case between orientation-selective mechanisms in highercortical areas involved in the processing of global stim-ulus properties (Poom, 2000; van der Zwan & Wenderoth,1995; Wenderoth & Johnstone, 1988).In this study, we measured TAEs for two types of

contour between textured areas, both of which elicitedthe percept of abutting or overlapping surfaces. One typeof contour demarcated two identical textures; hence, theedge was defined by aligned texture line endings only(SC; Figure 1). The other type of contour demarcated twotexture surfaces of different orientations; hence, there wasalso an orientation contrast across the edge (OC; Figure 2).TAEs have previously been observed using both Bsingle[SCs (Paradiso et al., 1989; van der Zwan & Wenderoth,1994, 1995) and gratings defined by SCs (Berkley et al.,1994; Bockisch, 1999; Montaser-Kouhsari & Rajimehr,2004). They have also been shown to interact with LCs(Berkley et al., 1994; Bockisch, 1999; Paradiso et al.,1989). Tilt aftereffects using texture edges defined byorientation contrast have not, to our knowledge, previ-ously been investigated. However, the existence of filterstuned to the orientation of such texture edges is stronglysuggested by the effect of adaptation to orientation-definedform on its perceived spatial location (Prins & Mussap,2001) and on contrast thresholds for its detection (Kwan &Regan, 1998).Stimulus presentation times in our study were sufficient

for contour orientation resolution to reach a maximum, asmeasured in an initial orientation discrimination experi-ment (Experiment 1). We wanted to test the encoding offully processed contours, and we took the time requiredfor maximal orientation discrimination as an indicator ofthe time required for maximal orientation encoding of our

Figure 1. An adapting stimulus for the SC condition. The contour

presented here is at 15 deg anticlockwise and is within a texture

that is at 90 deg to the contour. Test SC stimuli were very similar,

differing only in that the texture was always horizontal, regardless

of the angle of the SC contour.

Journal of Vision (2006) 6, 37–52 Hawley & Keeble 38

contours. This time was then used when presenting thecontours in the TAE parts of our study. In Experiment 2,we measured the intra-contour-type TAEs (adapt-testconditions OCYOC and SCYSC) over a range of adaptingangles to verify that both contour types displayed mis-perceptions of contour angle following adaptation. Thisexperiment allowed us to quantify the variation of TAEwith adapting angle for comparison with previous studiesand to locate the adapting angle for which maximum mis-perception of a test contour occurred. We also measuredthe adaptation effect of texture stimuli that did not containan adapting central contour (Figure 3) on the same texturecontour test stimuli. This control condition was conductedbecause the contour stimuli contained both a second-orderorientation cue in the form of the contours themselves andalso a first-order orientation cue in the form of the abut-ting textures that induced the SC and OC contours. Whilewe wished to observe the effect of adapting to the SC andOC contours, it became apparent that adaptation to thetextures themselves could also affect subsequent misper-ceptions of the test contour orientation. This effect wasquantified, and it was shown that the direct TAE observedwith the texture contours was not due to adaptation to theorientation of the composing textures.In Experiment 3, we measured the TAE at the maxi-

mal adaptation angle found in Experiment 2 (15 deg) forcross-adaptation conditions between OC and SC. Totaltransfer of adaptation effects between our matched SC andOC contours would indicate that the presence of orienta-tion contrast does not affect the final encoding of orien-tation of a texture surface edge, whereas nonexistent

transfer would point to mutually exclusive mechanismsfor the orientation encoding of the two texture contours.Although both types of contour contain aligned line end-ings, it could be that the presence of orientation con-trast across the edge is a Bstrong[ cue, which effectivelyBoverrides[ weaker mechanisms processing SCs. Incom-plete transfer of adaptation between contour types wouldsuggest limited shared sites of orientation encoding. Wealso measured the interaction of the texture contourswith an LC. An LC can be expected to elicit response incells tuned to the orientation of the contour in both V1and V2, and so a comparison of the adaptation effect ofLC compared to OC and SC will indicate the extent towhich the texture contours activate this population ofneurons.

Methods and results

Observers

Four observers participated in these experiments. Theauthors, SH and DK, were experienced psychophysicalobservers. Observers JI and AB were inexperienced ob-servers who were naive to the purpose of the experiment.All subjects had normal or corrected-to-normal vision.

Stimuli and apparatus

Stimuli were produced using a G3 PowerMac com-puter and were displayed on a ProNitron monitor that had

Figure 2. An adapting stimulus for the OC condition. The contour

presented here is at 15 deg anticlockwise, and the texture is at

T75 deg to the contour. Test OC stimuli were very similar, differing

only in that the texture was always at T75 deg to the vertical,

regardless of the angle of the OC contour. For test stimuli, the

order of the two texture orientations across the contour was

randomised, and for adapting stimuli the order was randomly

reallocated at each stimulus refresh (every 500ms).

Figure 3. A control-adapting stimulus. This example is the control

for the SC stimulus shown in Figure 1; texture is at the same

orientation in both stimuli, but here there is no contour demar-

cated by aligned line endings. A control stimulus for the OC

shown in Figure 2 could have texture at either of the two com-

posing texture orientations, and the orientation was randomly

reallocated from the two available orientations at each stimulus

refresh (every 500 ms).

Journal of Vision (2006) 6, 37–52 Hawley & Keeble 39

a pixel resolution of 30 pixels degj1 at a viewing distanceof 58 cm. All stimuli were presented within a circularaperture with a diameter of 20 deg, created by placing adark card over the monitor. The display was viewed in adark room, and the monitor edges and other cues to ver-tical and horizontal were not visible when attending thestimulus. Observers’ head position was stabilized by useof a head and chin rest.Two types of texture contour stimuli were used as

adapting and test stimuli. Textures were composed ofrandomly placed line elements, of lengths drawn from asquare distribution ranging from 2 to 25 arcmin. Linewidths were always 1 pixel, or 2 arcmin. Line elementswere presented at a Weber contrast of 50% on a uniformgrey background at 33 cdmj2. Contours passed throughthe central point of the display, which was marked by amaximum contrast fixation point during all phases ofstimulus presentation. A central fixation point was used tomaximize localization of the Badaptation[ effect withinretinotopically mapped V1 and V2. SCs demarcated twotexture areas of identical orientation and were definedonly by the alignment of line endings (Figure 1); if arandomly placed line element extended to the contour, itwas terminated at that location. This was achieved bycreating two full-screen textures and then splicing themtogether at the location of the contour, hence avoidingluminance artefacts that could result from anti-aliased lineendings. For SC-adapting stimuli, texture was alwaysoriented at 90 deg to the contour, but for SC test stimulithe texture had a fixed texture orientation of 90 deg re-gardless of the contour orientation so as to prevent textureorientation from cueing the test contour orientation. OCswere similarly defined by the alignment of line endings,and also by a change of texture orientation across thecontour (Figure 2). The two areas of texture were orientedat T75 deg relative to the OC for adapting stimuli, and atan absolute value of T75 deg to vertical for OC test stim-uli, with the order of the two texture orientations acrossthe contour being randomized. (Again, the use of fixedtexture angles for test stimuli meant that texture orienta-tion could not cue contour orientation.) The total orien-tation contrast across the contour was therefore 30 deg,which elicits a maximum perception of two distinct re-gions (Motoyoshi & Nishida, 2001). The high density ofline endings throughout both SC and OC stimuli meantthat an isolated cluster of line endings was unlikely tosignal the presence of the contour to observers. It wasintended that integration of local components on a largerscale should be required to indicate the presence of anSC. An LC was also used in some conditions. This was adark line of 2 arcmin (1 pixel) width presented at 50%Weber contrast on the same grey background. All contourtypes extended across the full diameter of the circularviewing aperture.To ascertain whether adaptation effects after viewing

OC and SC stimuli were due to adaptation to the contour

orientation, or to adaptation to the first-order orientationcue of the texture lines, two control texture stimuli werecreated. The possibility that the lines inducing an SC,rather than the SC itself, may be contributing to the ob-served TAE has not, to our knowledge, been previouslyconsidered. The control texture stimuli were matched tothe textures composing the SC and OC stimuli but did notcontain a central contour; textures continued uninterruptedacross the screen (Figure 3). The SC-matched control(SCC) consisted of unbroken texture at the same angle aswould have been used to induce an SC in any given stim-ulus presentation. The OC-matched control (OCC) couldhave one of two texture orientations, as the OC contourstimulus contained two texture orientations. In this case,the texture orientation is any single 500ms presentation pe-riod was chosen randomly from the two available orienta-tions. The SCC and OCC textures were used as adaptingstimuli in trials with SC and OC test contours, respectively.

Experiment 1: Time course of orientationdiscrimination for OCs and SCs

Experiment 1 measured orientation discriminationthresholds for OCs and SCs as a function of stimuluspresentation duration. This experiment was conducted tocompare the orientation resolution of the two contoursand to find the presentation duration required for maximalorientation resolution. A presentation duration that pro-duces maximal orientation resolution, presumed to equateto maximal low-level orientation encoding, would be usedfor the later TAE experiments. Subjects SH and DK wereused for this experiment.The method of constant stimuli was used to present con-

tours at a range of orientations close to vertical. Contoursorientations ranged from j3 to 3 deg, at intervals of0.75 deg. Observers indicated, by means of a key press,whether the presented contour was tilted clockwise oranticlockwise. Observers were free to respond in their owntime, with response initiating the next stimulus presenta-tion. Contours were presented for 50, 100, 200, 500,or 1000 ms. Trials were blocked for contour type andpresentation duration, with a total of 360 observationscompleted for each combination of contour type andpresentation time. Probit analysis (Finney, 1971) was usedto fit psychometric functions to the data, giving theposition of perceived vertical (PV) and the standarddeviation of the PV, also called the orientation discrim-ination threshold (the orientation yielding a performanceof 84% correct for a PV of 0).

Results of Experiment 1

Both subjects showed a bias in PV, with observerSH judging contours to be tilted clockwise and observerDK tilted anticlockwise. Biases for both observers wereG1 deg and did not differ with contour type or presentation

Journal of Vision (2006) 6, 37–52 Hawley & Keeble 40

time. These biases can be a true misperception of PV orbecause clockwise and anticlockwise responses were al-ways indicated by the same key presses, a bias in response.However, as the bias was constant for each observer, thisdoes not affect our later comparisons of TAE.The results for the orientation discrimination threshold

are displayed in Figure 4. Both observers showed a de-crease in threshold with time. Thresholds initially dropsharply but then plateau, showing little further decreasebeyond presentation times of 300Y400 ms. At presentationtimes beyond 100 ms, threshold values are not signifi-cantly different between the contours. As we wished toobserve adaptation effects for borders that were maxi-mally encoded, we subsequently used presentation timesof 500 ms for both contour types when measuring PVbefore and after adaptation and a corresponding presenta-tion frequency of 2 Hz when refreshing contour stimuliduring adaptation. Shorter stimulus presentation times maygive larger TAEs (Wolfe, 1984), but this would defeatthe purpose of our experiment, to compare the orienta-tion encoding of the two types of second-order contourat a point in time when they are maximally encoded.In addition, the misperception of orientation observed us-ing short (G100 ms) stimulus presentations may in fact besubserved by different mechanisms to those underlyingthe misperception seen with the longer stimulus presen-tations (Wolfe, 1984) that are more usually used in TAEstudies.

Experiment 2: Tuning curves for the TAEusing SCs and OCs

In Experiment 2, we measured the TAE for OCs andSCs at adapting angles from 0 to 90 deg. The experimentwas replicated using the texture control stimuli, OCC andSCC, as adapting stimuli while testing with the matchedcontour stimuli. Observers SH, DK, and AB participatedin Experiment 2.Each trial consisted of three phases: preadaptation

measurement of PV, adaptation, and postadaptation mea-surement of PV. During the adaptation phase, the adapt-ing stimulus was presented for 60 s, with the stimulusrefreshed at a rate of 2 Hz. The refresh changed the in-dividual lines composing the textures on either side of theadapting contour, without changing the contour orienta-tion. Rather than the refresh occurring instantaneously,stimuli were presented for 400 ms and interleaved with a100-ms blank screen. The blank screen itself was not per-ceived; the impression was of a stimulus onset at 500-msintervals. Both the change of texture local compositionand the nature of the stimulus onset served to increasethe saliency of the adapting contour, as it was effectivelyBreformed[ at rate of 2 Hz. Additionally, changing thelocal texture composition limited adaptation caused bylocal luminance differences while maintaining the adap-tive influence of the contour under investigation. Subjectswere instructed to direct their gaze at the central fixationspot at all times but to attend to the entire contour, thatis, to maintain awareness of the contour presence duringthe prolonged initial adaptation. TAE was measured usingadapting contours presented at 5Y20 deg at intervals of5 deg, and then to 90 deg at intervals of 10 deg. Threemeasurements were made for each adaptation angle pre-sented in the clockwise direction, and three measurementswere made in the anticlockwise direction, except for ob-server SH who made 5 TAE measurements in each direc-tion for the SCYSC and OCYOC trials.The pre- and postadaptation measurements of PV used

a single staircase method. In an experimental session,both PV measurements started with a contour presentedto the same side of vertical, which was nominally the di-rection of the adapting stimulus. On the first trial of eachstaircase, the orientation of the test contour was chosenrandomly to be 4, 5, 6, or 7 deg from vertical. On sub-sequent trials, the contour was rotated in the directionopposite to that reported by the observer, such that pre-sentations converged to PV. The contour was rotated by2 deg until the first reversal in observer response, 1 deguntil the second reversal, and 0.5 deg thereafter until10 reversals had been made. Contour orientations at thelast six reversals were averaged to find the subject’s PV.During the preadaptation measurement phase, observerresponse initiated the next presentation. In the postadap-tation measurement phase, observer response initiated a6 s period of Btop-up[ adaptation, which was of the sameform as the initial adaptation. The next test presentationfollowed after a 500 ms blank screen. The preadaptation

Figure 4. Orientation discrimination thresholds for the two texture

contours used in the TAE experiments. The data sets are fit with a

function of the form y ¼ c þ xm for illustrative purposes only, with

data points weighted by their standard errors (error bars shown).

Journal of Vision (2006) 6, 37–52 Hawley & Keeble 41

PV was subtracted from the postadaptation PV to give themagnitude of the TAE, and the absolute value of the TAEwas averaged over clockwise and anticlockwise trials ateach adaptation angle. This removed any bias in observerjudgement caused by the direction from which the contourapproached PV, as the approach direction was the samefor the pre- and postadaptation measurements.The use of a single staircase and the length of initial and

Btop-up[ adaptation periods were designed to limit the totaladaptation time while maintaining adaptation effects.Prolonged periods of adaptation have long-term measur-able aftereffects (Wolfe & O’Connell, 1986), a problemwe wanted to be sure of avoiding so as not to require longbreaks between experimental sessions. However, if adap-tation is insufficient, TAEs are not seen. Our experimentaldesign resulted in a ratio of top-up adaptation to non-adaptation time of approximately 75%, depending onthe delay before observer response (as this determined theinterval between test stimulus offset and the start of thenext top-up period). This ratio was found by Wolfe andO’Connell (1986) to be sufficient to induce adaptationusing luminance-defined contours, not just maintain it.Pilot work confirmed that this ratio was also sufficient toinduce adaptation using SC and OC contours. Each trialrequired roughly 20 top-up adaptations, which, togetherwith the initial 60 s of adaptation, made a total adaptationtime of 3 min. The TAE on LCs after this duration ofadaptation was found to drop off rapidly by Wolfe andO’Connell (1986), returning to less than a quarter of itsmaximum after 5 min, with minimal long-term adaptation.Accordingly, our observers were required to rest for 5 minbetween trials. As a further precaution against cumula-tive adaptation effects, clockwise and anticlockwiseadaptation stimuli were interleaved.

Results of Experiment 2

Graphs of TAE versus adaptation angle are shown forSCs (adapt-test condition SCYSC; Figure 5) and OCs(adapt-test condition OCYOC; Figure 6). It can be seenthat both contour types show TAE curves with a positivepeak at approximately 15 deg and a negative peak closeto 70 deg, typical of luminance contrast contours (e.g.,Gibson & Radner, 1937) and also previously shown withSCs (van der Zwan & Wenderoth, 1995). Figures 5 and 6also show the results of adaptation using matched texturecontrol stimuli (for example stimulus, see Figure 3),which gave negligible TAEs for SC and OC test contoursin the region of 0Y30 deg. The results of curve fitting, asdetailed in the following paragraphs, suggest that adapta-tion to the texture orientation, as composes the control-adapting stimuli, cannot account for the TAE seen in thisregion for the SCYSC and OCYOC conditions. However,the texture controls do elicit a significant negative TAEpeaking at ¨70 deg. This strongly suggests that the neg-ative TAE seen for the SC and OC contours at approx-imately 70 deg can also be attributed to adaptation to the

first-order orientation cue of the textures and is not aconsequence of adaptation to the contours themselves.The data sets for SCs and OCs were fit with a difference

of Gaussian (DoG) function:

y ¼ Axejx2

2B2 þ C xj90ð Þejðxj90Þ2

2D2

!=2; ð1Þ

where A and C are the mean amplitudes of the twoGaussian contributions (which are of opposite polarities),and B and D are their respective standard deviations.Curve fitting was achieved using Kaleidagraphi data

Figure 5. Tuning curves of the TAE in a subjective test contour after

adaptation to an SC or a matched texture control (SCC). Error bars

are standard errors. The full data sets from each observer (DK, AB,

and SH) are fit with DoG functions (Equation 1).

Journal of Vision (2006) 6, 37–52 Hawley & Keeble 42

analysis software, which uses the LevenbergYMarquadtalgorithm (Press, 1992), and data points were weighted by1/SE2. The DoG function gave a good fit to the data, withPearson’s correlation coefficient (BR values[) 9.95 in allcases (Table 1; SCYSC and OCYOC). The first Gaussianrepresents the adapting effect of the contour, a second-order cue to orientation. The second Gaussian is taken torepresent the first-order adapting effect of the texturelines, which are oriented at 90 deg to the contour (thisvalue is exact for SC and an average for OC as textureorientation had two values). Linear summation of two pop-ulations of activity has been used previously to model

cross-orientation interaction (Blakemore et al., 1970), andthe linear summation of Gaussian functions in particu-lar has been used to model the interaction between first-and second-order orientation processing (Skillen, Whitaker,Popple, & McGraw, 2002). While we do not assume va-lidity of the Gaussian function in representing underlyingadaptive processes, both the strength of the fit and the out-come of various manipulations of the function parametersdescribed suggest that the summation of two functions rep-resenting the first- and second-order orientation adaptationcues present in our stimuli provides a robust descriptionof our data.In Equation 1, the second Gaussian is at a fixed offset

of 90 deg from the first. When the Boffset[ of the sec-ond Gaussian was set as a free parameter, it converged to90 T 5 deg in all cases, and the fit of the function to thedata (as shown by R values) did not systematically im-prove, suggesting that the source of the attraction is in-deed located at an angle of 90 deg to the source of therepulsion effect. Secondly, we explored holding the twoGaussian bandwidths equal, that is, B = D. While thismodification of the function was not detrimental to the fitof SC data, it resulted in lower R values for fits to orien-tation contrast data. When the two bandwidths were al-lowed to vary independently, the bandwidth allocated tothe OC, texture (D, centered at ¨70 deg), was considerablygreater than that allocated to the OC contour itself (B, cen-tered at ¨15 deg).The difference between the two bandwidths was much

smaller for the SC condition, hence holding the contourand texture bandwidths equal made little difference to thestrength of the fit. We would suggest that the larger band-width for adaptation attributed to the first-order cue of thetexture in the OC stimulus was because the texture at anygiven spatial location was randomly presented at one oftwo different orientations that were 30 deg apart. Thispossibility is supported by the finding of smaller ampli-tudes for the first-order texture orientation Gaussian(parameter C) in the OC case compared to the SC case,that is, the adaptive effect of the OC texture was less strongbut more Bspread out[.The second two row sections of Table 1 describe fits to

the control conditions (SCCYSC and OCCYOC). As wellas fitting the data with the DoG function, it seemed log-ical to fit the data with just the negative Gaussian portionof the DoG, as the source of the hypothesized positiveportion (the SC or the OC contour) is not present in thecontrol stimulus. For observer AB, the fit achieved withthe single, negative Gaussian was nearly as good as theDoG fit. Note that fitting with a single Gaussian meansthat there are less free parameters and so the goodness offit is likely to be decreased. However, for observer DK,the fit with the negative Gaussian was very poor and thereason for this is not known.Finally, we looked at the effect on the curve fit of sub-

tracting the control data (SCCYSC and OCCYOC) fromthe SCYSC and OCYOC data, which should this time

Figure 6. Tuning curves of the TAE in an orientation contrast test

contour after adaptation to an OC or to a matched texture control

(OCC). Error bars are standard errors. The full data sets from each

observer (DK, AB, and SH) are fit with DoG functions (Equation 1).

Journal of Vision (2006) 6, 37–52 Hawley & Keeble 43

leave just the positive portion of the curve (last two rowsections of Table 1). The data were therefore fit with thesingle, positive Gaussian as contained in DoG, as well aswith the DoG itself. The single Gaussian gave a good fitto the data in most cases.In summary, and to return to the main finding of

Experiment 2, while OCYOC and SCYSC gave a signifi-cant positive Gaussian (parameter A in Table 1), thecontrol conditions OCCYOC and SCCYSC did not. Thisindicates that the direct TAE seen peaking at an adaptiveangle of ¨15 deg for SC and OC stimuli cannot be ac-counted for by adaptation to the texture orientation but iscaused by the second-order contours themselves.

Experiment 3: Cross-adaptation effects for SCs,OCs, and LCs

In Experiment 3, we measured the TAE at an adapta-tion angle of 15 deg, the peak of the SC and OC adap-tation effect found in Experiment 2, for cross-adaptationconditions between SC and OC. This investigated the

extent to which the two types of second-order contouractivate the same set of orientation-selective neurons. Inaddition, several adapt-test contour combinations amongSC, OC, and a luminance-defined contour (LC) were tested,similarly, to investigate the overlap of orientation encod-ing between the two second-order contours and a first-ordercontour. Observers SH, DK, AB, and JI participated in thisexperiment.The procedure was identical to that of Experiment 2.

All subjects were tested with the adapt-test combinationsof subjectiveYsubjective (SCYSC), orientation contrastYorientation contrast (OCYOC), and matched texture controls(SCCYSCC and OCCYOCC) to ascertain that they perceivedTAEs attributable to the texture contours. All possibleadapt-test combinations of LC, SC, and OC contours weretested, allowing a comparison of (i) cross-adaptation be-tween texture contours (OCYSC, SCYOC) compared withthe within-contour conditions (OCYOC, SCYSC); (ii) adap-tation to an LC compared to adaptation to texture con-tours, on texture test contours (LCYOC versus OCYOC andLCYSC versus SCYSC); and (iii) adaptation on a lumi-nance test contour by all three adapting contours (LCYLC,

Adapt-

test

condition Observer

Data

range

(deg) Type of fit

A

(repulsive

mean

height)

B

(repulsive SD)

C

(attractive

mean

height)

D

(attractive SD)

R

(Pearson’s

correlation

coefficient)

SCYSC SH 0Y90 DoG 0.235 T 0.045 14.174 T 1.672 0.313 T 0.032 18.475 T 1.1272 .991

AB 0Y90 DoG 0.378 T 0.093 14.137 T 2.007 0.203 T 0.076 16.140 T 3.552 .971

DK 0Y50 DoG 0.293 T 0.088 14.456 T 5.617 0.146 T 0.430 25.363 T 25.068 .959

OCYOC SH 0Y90 DoG 0.400 T 0.067 19.876 T 1.012 0.126 T 0.014 36.921 T 5.348 .981

AB 0Y90 DoG 0.538 T 0.045 16.258 T 1.1658 0.153 T 0.049 18.934 T 3.409 .979

DK 0Y50 DoG 0.530 T 0.179 18.830 T 3.847 0.111 T 0.127 41.610 T 20.620 .973

SCCYSC AB 0Y90 DoG 0.045 T 0.024 j35.144 T 8.075 0.252 T 0.046 22.992 T 2.496 .970

Negative Y Y 0.258 T 0.045 19.338 T 1.577 .927

DK 0Y50 DoG 0.153 T 0.122 23.602 T 15.461 0.097 T 0.158 39.465 T 21.498 .938

Negative Y Y 0.594 T 2.103 17.105 T 10.443 .434

OCCYOC AB 0Y90 DoG 0.118 T 0.079 25 T 8.279 0.090 T 0.044 43.460 T 11.113 .800

Negative Y Y 0.101 T 0.027 27.633 T 3.579 .759

DK 0Y50 DoG 0.253 T 0.081 28.316 T 21.298 0.204 T 0.341 41.805 T 17.021 .913

Negative Y Y 0.332 T 0.299 21.786 T 4.801 .600

SCYSC

minus

SCCYSC

AB 0Y90 DoG 0.514 T 0.214 9.406 T 2.223 j0.052 T 0.039 31.512 T 13.656 .913

Positive 0.478 T 0.184 10.458 T 2.284 Y Y .687

DK 0Y50 DoG 0.327 T 0.195 9.998 T 3.175 j830.220 T 1.038 e + 07 j8.026 T 40169 .904

Positive 0.327 T 0.195 9.998 T 3.175 Y Y .903

OCYOC

minus

OCCYOC

AB 0Y90 DoG 0.548 T 0.065 16.142 T 1.544 0.148 T 0.168 10.311 T 6.179 .968

Positive 0.548 T 0.066 16.122 T 1.542 Y Y .958

DK 0Y50 DoG 0.381 T 0.293 14.496 T 5.731 j0.018 T 0.054 70.048 T 244.11 .964

Positive 0.447 T 0.090 15.936 T 2.225 Y Y .937

Table 1. Parameters from the DoG curve fit to the TAE data. Amplitude and standard deviation of the repulsive and attractive parts of the

curve fit of Equation 1 (DoG), the first ‘‘positive’’ Gaussian only (positive) or the second ‘‘negative’’ Gaussian only (negative). ‘‘Y’’

indicates parameters that are not applicable to the fit.

Journal of Vision (2006) 6, 37–52 Hawley & Keeble 44

OCYLC, and SCYLC). Adapt-test conditions were inter-leaved, and clockwise and anticlockwise adaptations ateach orientation were paired within conditions and werealternated as before. TAEs were explored using within-subjects (repeated-measures) analysis of variance (ANOVA)to test for the significance of adapting contour and testcontour factors. Where this resulted in a multiple com-parison of a single adapt-test condition being made, theappropriate Bonferroni correction was made when test-ing for significance at the 95% level: two comparisons,p = .0250; three comparisons, p = .0167; four comparisons,p = .0125.

Results of Experiment 3

All adapt-test combinations, with the exception of adap-tation to the matched texture controls (OCCYOC andSCCYSC), were significantly greater than zero. Figure 7shows the magnitude of the TAE for the original intra-contour-type conditions, OCYOC and SCYSC, and theirrespective texture controls. A paired t test (i.e., one-waywithin-subjects ANOVA with two levels) of the OCYOCand SCYSC conditions showed them to be significantlydifferent (1.85 versus 1.04; p = .006), with the OCYOCTAE on average 0.81 deg greater than that for SCYSC.OC contours must be stronger adaptors of orientation-selective neurons in V1/V2 than SC contours, or must bemore susceptible to adaptation effects, or both. Thecontour TAEs were not explained by the adaptive effectof the textures, confirming that the subjects perceived andadapted to the contours themselves as previously shownin Experiment 2. Hence, it is demonstrated that the TAEis larger for OC contours, which contained an orientationcontrast cue, than for SC contours.Figure 8 shows the results of cross-adaptation between

the two texture contour types, and also the initial intra-contour-type conditions for comparison. On first inspec-tion, there is no apparent difference in TAE between the

two cross-adaptation conditions (OCYSC versus SCYOC),but a two-way (adapting contour, test contour) within-subjects ANOVA reveals that there is a strong effect ofadapting contour (F = 60.59, p G .005). Planned orthog-onal pairwise comparisons showed that this was due toa strong increase for OCYSC compared to the SCYSCcondition (1.49 versus 1.04; p = .00) and a nonsignificantdecrease for SCYOC compared to the OCYOC condition(1.55 versus 1.85; p = .052). The effect of test contour wasalso significant (F = 22.24, p G .05), with OC test contoursdisplaying larger TAEs. This was mainly due to the strongincrease for SCYOC compared to SCYSC (1.55 versus1.04, p = .005), as the increase for OCYOC compared toOCYSC was nonsignificant (1.49 versus 1.85, p = .072).Hence, comparison of the cross-adaptation conditions withthe original conditions adds weight to the initial sugges-tion that OCs are stronger adaptors and are more sus-ceptible to adaptation effects than SCs.Next we compared the TAE induced on an OC after

adapting to an LC and after adapting to an OC contour(i.e., the within-contour-type condition). The equivalentcomparison was made for an SC (Figure 9). Two separatepaired t tests showed that there was a trend for adaptationby an LC to produce larger effects on the texture contoursthan adaptation by the same texture contour, but neithercomparison reached significance (LCYOC versus OCYOC,2.09 versus 1.85, p = .22; and LCYSC versus SCYSC, 1.92versus 1.04, p = .059).Figure 10 shows the TAEs induced on a luminance

test contour (LC) after adapting to the three contours(LC, OC, and SC). A one-way within-subjects ANOVAshowed no difference between the three conditions, sug-gesting that OC and SC were as effective at adaptingthe population of neurons encoding the LC orientationas the LC contour itself was. As OC and SC contours are

Figure 7. TAE magnitudes at an adaptation angle of 15 deg for

intra-contour-type adapt-test conditions (orientation contrastY

orientation contrast, OCYOC; subjectiveYsubjective, SCYSC) and

their respective texture control conditions (OCCYOC; SCCYSC).

Error bars show standard errors.

Figure 8. TAE at an adaptation angle of 15 deg for cross-contour-

type conditions (subjectiveYorientation contrast, SCYOC; orientation

contrastYsubjective, OCYSC) compared to the intra-contour-type

conditions (OCYOC; SCYSC). Error bars show standard errors.

Journal of Vision (2006) 6, 37–52 Hawley & Keeble 45

second-order orientation signals and LC is a first-orderorientation signal, this result was slightly surprising.Hence, we also compared the conditions OCYOC versusOCYLC and SCYSC versus SCYLC in two separate pairedt tests. This comparison checked if the texture contourswere as effective at adapting neurons encoding a first-order luminance-defined orientation signal as they wereat adapting neurons sensitive to the second-order tex-ture contour orientation. We found that OCYOC gave agreater TAE than OCYLC, but that SCYSC gave a smallerTAE than SCYLC, and that neither comparison reachedthe required significance level (OCYOC versus OCYLC,mean difference ¼ 0:875, p = .026; SCYSC versus SCYLC,mean difference ¼ 0:298, p = .499). Hence, in this com-parison, there was no significant difference in the abilityof the texture contours (OC and SC) to adapt neurons en-coding second-order texture contour orientation and first-order LC orientation.

Discussion

The results of the present study confirm previousobservations of a TAE using SCs (Berkley et al., 1994;Bockisch, 1999; Paradiso et al., 1989; van der Zwan &Wenderoth, 1994, 1995) and of their interaction with LCs(Berkley et al., 1994; Bockisch, 1999; Paradiso et al.,1989). We also found strong TAEs for our OC, a type ofcontour that to our knowledge has not been investigatedin this way before. This result is compatible with thepreviously demonstrated orientation-tuned detectionthreshold elevation after adaptation for orientation texture-defined gratings (Kwan & Regan, 1998). Another novelresult of this study is the finding of a stronger adaptiveeffect of texture edge contours with OC than without (SC).This answers our original question, showing that theaddition of an orientation contrast cue across a textureedge strengthens the encoding of orientation at the levelof V1/V2. This finding is not entirely to be expected,as the two second-order contours had comparable orien-tation encoding in terms of orientation discriminationthresholds. We did not seek to use contour stimuli matchedfor salience, a factor that is known to influence the magni-tude of the TAE (Berkley et al., 1994). Rather, we soughtto differentiate the relative salience of two kinds of tex-ture contours. The contours were identical apart from theattribute under investigation, orientation contrast. Hence,differences in the adaptive strength of our two texturecontours are an effect of orientation contrast only.In addition to the primary aim of the experiment, we also

further investigated the encoding of second-order contourorientation by comparisons with a first-order luminance-defined contour. We found that adaptation by an LCproduced a TAE with OC and SC test contours, evidenceof shared sites of orientation encoding between first- andsecond-order orientation signals. Note that adaptation byan LC induced slightly larger TAEs in SC and OC testcontours than SC and OC adaptors themselves (althoughthe difference was not significant), showing that anorientation signal defined by luminance contrast is equallyeffective at adapting the population of neurons encodingthe orientation of the texture contours. Both Berkley et al.(1994) and Paradiso et al. (1989) also found that adap-tation by an LC could induce equal or larger TAEs in anSC than adaptation by the SC itself.However, more importantly, this study found no differ-

ence in the effectiveness of the LC, OC, and SC adaptorson an LC test contour. Likewise, there was no differencebetween OC and SC adaptors on an LC test contourcompared to the OCYOC and SCYSC conditions. This is asurprising result when it is considered that luminance-defined orientation signals are first detected in V1, and itis generally accepted that the TAE seen with luminance-defined contours results from altered orientation-selectiveresponse in V1 and V2. For example, Paradiso et al.(1989) found that their SCs induced much weaker TAEs

Figure 10. TAE at an adaptation angle of 15 deg induced on a

luminance test contour by an LC (LCYLC), by an SC (SCYLC),

and by an OC (OCYLC). Error bars show standard errors.

Journal of Vision (2006) 6, 37–52 Hawley & Keeble 46

Figure 9. TAE at an adaptation angle of 15 deg for adaptation by

an LC (luminanceYorientation contrast, LCYOC; luminanceY

subjective, LCYSC) compared with adaptation by texture contours

(OCYOC; SCYSC) on test texture contours. Error bars show stan-

dard errors.

in LCs than those induced by LCs themselves. Theyexplained their results in terms of the relative populationsof V1 and V2 that might encode the two contours; whilethe LC could activate all orientation-tuned cells, includ-ing those that signaled SC orientation, they hypothesizedthat the SC could only activate a subset of the cells thatencoded the LC orientation. Hence, they reasoned thatthe effect of adaptation by an SC would therefore be lessnotable within the entire population of cells signaling theorientation of a luminance test contour.Although a few studies have found a small V1 response

to SCs (Lee & Nguyen, 2001; Ramsden, Hung, & Roe,2001; Redies, Crook, & Creutzfeldt, 1986; Sheth, Sharma,Rao, & Sur, 1996), it is more usually found that a robustresponse to both subjective and texture orientation con-trast contours is first seen in V2 (Baumann et al., 1997;Leventhal et al., 1998; Sheth et al., 1996; von der Heydt& Peterhans, 1989; von der Heydt et al., 1984). LCsshould therefore have a greater adaptive effect than OCand SC on luminance test contours, according to the ex-planation put forward by Paradiso et al. (1989). However,our results suggest that all three contours are equallyeffective at activating V1/V2 during prolonged viewing.A plausible explanation for this is that OC and SC ac-tivate orientation-selective neurons in V1 by feedbackfrom V2 or higher cortical areas and thereby indirectlyproduce adaptation in V1. This line of reasoning and apossible explanation for the difference between the resultsof this study and those of Paradiso et al. (1989) is dis-cussed more fully later.

The TAE "tuning curve"

Both SC and OC texture contours displayed typicalBS[-shaped tuning curves (cf. van der Zwan & Wenderoth,1995), with a Brepulsive[ peak centered at ¨15 deg andan Battractive[ peak centered at ¨70 deg. Investigationof TAEs and TIs caused by stimuli with multiple axes ofsymmetry has suggested that the attractive effect, oftenattributed to Bglobal[ properties of the stimulus ratherthan local orientation signals, may be due to adapta-tion to the orthogonal axis of symmetry in line or gratingstimuli (Johnstone & Wenderoth, 1989; Smith, Wenderoth,& van der Zwan, 2001; Wenderoth & Johnstone, 1987;Wenderoth, Johnstone, & van der Zwan, 1989; Wenderoth,van der Zwan, & Johnstone, 1989).We propose that the Battractive[ effect for our stimuli

can be explained by adaptation to the first-order cue oftexture orientation, and that it is unnecessary to hypothe-size adaptation to an axis of symmetry orthogonal to thesecond-order contours (OC and SC). Rather than the testcontour being attracted in the direction of the adaptingcontour, at large adapting contour angles, it appears tohave been repulsed by the texture orientation. Hence, bothportions of the curves shown in Figures 5 and 6 can be at-tributed to repulsive effects. However, we cannot rule out

the possibility that there was a small adaptive effect ofthe contour axis of symmetry that was not observed dueto the strong adaptation effect of the texture. Indeed, vander Zwan and Wenderoth (1995) observed an attractiveeffect peaking at 75 deg for their SCs, which consisted ofoffset concentric circles, and hence contained no globalfirst-order orientation signal at all.The good fit to our data obtained using a DoG func-

tion, which sums adaptation effects rather than dis-carding weaker influences, is in agreement with someother studies that have used multiple adapting orientations(Magnussen & Kurtenbach, 1980b; Wenderoth, van derZwan, & Williams, 1993). An earlier study had suggestedthat the adaptor presented closest to the test angle (ver-tical test in our case) overrides the influence of poten-tial adaptors presented at larger angles (Wenderoth &Curthoys, 1974). Because the two adaptation effects inour stimuli were at 90 deg to each other, our stimuli arenot ideally suited to examining this issue. Suffice to saythat the excellent fits achieved with DoG functions werenot improved by fitting the 0Y45 deg and 45Y90 degportions of the curve separately, as would be necessaryif the contour and texture adaptation influences weremutually exclusive.

The relative adaptive strengths ofSC, OC, and LC contours

We found considerable variations in TAE across sub-jects, both in the absolute magnitude and in the relativeeffectiveness of the different contour types. Similarvariations have been noted before (Berkley et al., 1994),and these may reflect individual differences in perceptionof the contours. It could also be of particular relevancethat the salience of SCs is decreased by strong fixation(Bradley & Dumais, 1984; Frisby & Clatworthy, 1975;Kanizsa, 1976; Soriano, Spillmann, & Bach, 1996).Indeed, all our observers reported difficulty in Bseeing[the SCs during both the initial, and the shorter Btop-up,[adaptation phases. The detrimental effect of fixation onSC visibility could be because fixation is limiting aware-ness of the global context that induces the contour.Awareness of global stimulus properties is known toheighten the later parts of neural response in V1 and V2(Ito & Gilbert, 1999; Lamme, Zipser, & Spekreijse, 1998;Lee & Nguyen, 2001; Lee, Yang, Romero, & Mumford,2002; Super, Spekreijse, & Lamme, 2001; Tanaka, 2001).If we accept that the direct TAE is an aftereffect oforientation-selective activation in V1 and V2 (Blakemoreet al., 1970; Carpenter & Blakemore, 1973; Magnussen &Kurtenbach, 1980a, 1980b; Morrone et al., 1982; Wenderoth& Johnstone, 1987), then variability in observers’ awarenessof contours and the subsequent strength of contourrepresentation in these cortical areas could explain someof the variability in TAE magnitudes.

Journal of Vision (2006) 6, 37–52 Hawley & Keeble 47

Both the frequent change of the local composition ofour contour-inducing textures, and also explicit instruc-tions to observers, were intended to heighten awareness ofthe contours despite fixation. However, our observers mayhave maintained awareness of the contours to greateror lesser extents. Our reasoning that stimulus awarenesscould affect the encoding of our texture contours findssupport in the work done by Montaser-Kouhsari andRajimehr (2004). They found that the strength of the TAEinduced by an SC was inversely related to the attentionaldemands of an irrelevant task that was performed by ob-servers during the adaptation period. It was concluded thatthe attentional resources available for processing the SCaffected the strength of adaptation. Similarly in the pres-ent study, reduced awareness of the contour could limit itsadaptive strength. OC contours might be hypothesized tobe less affected by attentional resources, as the preatten-tive nature of local orientation contrasts would continu-ously signal the presence of this type of contour.The decreased visibility with time of SCs during the

initial and top-up adaptation phases may have also con-tributed to the finding of weaker adaptation by SC con-tours than by OC contours. Consider that the magnitudeof the TAE is determined by the relative contrast of theadapting and test stimuli when they are defined by lu-minance contrast (Parker, 1972). This principle also holdsfor positional adaptation (Whitaker, McGraw, & Levi,1997) and detection threshold elevation (Snowden, 1994),and it can presumably be accounted for in terms of therelative activities of the population of neurons activatedby the adapting and test stimuli that are defined by thesame attribute. An SC that has perceptually weakenedduring adaptation would be less effective than a brieflypresented SC in eliciting a neural response; the contoureffectively has lower contrast. Note that we are suggestingthat the decreased visibility of the SC is caused by a de-crease in the top-down influences on the contour vis-ibility, not by adaptation to the orientation signal. Hence,even if brief presentations of OC and SC contours wereequally effective in activating orientation-selective neu-rons in V1/V2 (as also suggested by our finding of equalorientation discrimination thresholds at short presen-tations), decreasing visibility of SC contours during theadaptation phase would result in lower TAEs for SCadaptor than for OC adaptor conditions. The differencebetween the two contours as test stimuli is harder to ex-plain, as OC test stimuli displayed larger TAE than SCdespite being the Bstronger[ contour. However, the dif-ference was only significant when SC was the previouslypresented adaptor, that is, SCYOC versus SCYSC. It maybe that interspersed presentations of OC increased orBrejuvenated[ the effectiveness of the SC Btop-up[adaptations.Decreased visibility with increasing presentation time is

certainly not the only possible explanation for the weakeradaptive effect of SCs when compared to OCs. SCs of the

type used here are widely accepted to activate a feedfor-ward process whereby line endings are detected in V1,and aligned responses input to orthogonal Bintegrating[cells in V2 (Peterhans & von der Heydt, 1989). OCs couldactivate an additional feedforward process mediated by aseparate population of neurons integrating local orienta-tion contrast. Possible mechanisms are the Blinking[ oflocal orientation singularities (Li, 2000), or a second stageof filtering by second-order filters that detect the region oforientation contrast (Bergen & Landy, 1991; Graham,1994; Malik & Perona, 1990; Popple, 2003; Wolfson &Landy, 1999).However, the equal adaptive strength of OC and SC

contours in V1/V2 compared to a first-order LC signalwould be difficult to explain by feedforward mechanismsonly. Our texture contours only give a second-order cueto contour orientation, which is likely to be first detectedinV2 (Baumann et al., 1997; Leventhal et al., 1998; Sheth et al.,1996; von der Heydt & Peterhans, 1989; von der Heydtet al., 1984). However, it appears that the representa-tion of the texture contours and an LC is very similar inits locus and strength within V1/V2. To adapt the sameorientation-tuned neurons as an LC contour, the repre-sentation of SC and OC contours may therefore rely onfeedback to some extent, possibly from V2 to V1 andalso from higher cortical areas to both V1 and V2. Thisfeedback appears to be sufficient to equate the SC/OCorientation-encoding response in V1/V2 with that causedby feedforward mechanisms responding to luminancecontrast. It is slightly surprising that the relative strengthsof the LC, OC, and SC adaptors could not be differ-entiated in Experiment 3, given that there were significantdifferences between OC and SC adaptors in Experiment 2.An alternative explanation for LC, OC, and SC adaptationequivalence in terms of a common site of orientation en-coding beyond V2 might be viable but would not be com-patible with the long-standing evidence that the first-order,luminance-defined TAE is generated in V1/V2.Orientation discrimination thresholds for our texture

edge contours (OC and SC) decreased with increasingpresentation time, reaching a plateau in the region of400Y1000 ms. Similar results have been reported beforefor SCs, whereas LCs or contours containing a luminancecue attain threshold plateaus for presentation times ofunder 100 ms (Westheimer & Li, 1996, 1997). Prolongedprocessing may point to an involvement of iterative feed-forward and feedback interactions (e.g., Lamme &Roelfsema, 2000; Lamme, Super, & Spekreijse, 1998;Lee & Mumford, 2003) in generating the texture contourpercept. The proposal has been made before that V1 andV2 responses to SCs (Lee & Nguyen, 2001) and textureorientation contrast contours (Lee, Mumford, Romero, &Lamme, 1998) have input from delayed modulation dueto feedback from higher areas that encode scene contoursin a cue-invariant fashion. Indeed, V1 and V2 recordingsin macaque show late modulation both in response to

Journal of Vision (2006) 6, 37–52 Hawley & Keeble 48

figures defined by orientation contrast (Borthogonal[ disc)and to matched subjective figures (Bparallel[ disc) similarto our OC and SC stimuli (Marcus & Van Essen, 2002).These findings support a possible role for feedback ingenerating the final orientation encoding of our SC andpossibly also OC contours.Both the SCs and the OCs were constructed so as to

resemble abutting or overlapping texture surfaces. Hence,we could expect the stimuli to be more salient than simpleline grating stimuli (such as used by Paradiso et al., 1989)to higher cortical areas that are responsive to figural sur-faces and occlusion percepts (Baylis & Driver, 2001; Grill-Spector, Kourtzi, & Kanwisher, 2001; Gulyas, Cowey,Heywood, Popplewell, & Roland, 1998; Kourtzi, Tolias,Altmann, Augath, & Logothetis, 2003) despite being iso-lated contours. Feedback from higher areas could there-fore be more active in modulating low-level response toour texture contours than for some other versions of an SC.As it was found that the orientation contours were

stronger adaptors than SCs, in the within-texture contourscomparison at least (Experiment 2), either the perception ofOC contours is less reliant on feedback and hence lesssusceptible to fading than the perception of SC contours, orfeedback is more robust when adapting to OC contours.Additional feedforward mechanisms for OC contours werediscussed previously. The alternative that OC contourscould elicit more robust feedback than SC could be theresult of the attentional properties of orientation singu-larities (Joseph & Optican, 1996; Nothdurft, 2002) orbecause of the first-order difference between the textureseither side of the OC contour. In the SC condition, thegeneration of a feedback input that could increase con-tour salience relies on the contour already being Bknown[by the visual system, as there is no difference in texturesurfaces either side of the contour.

Conclusion

To conclude, adaptation to orientation contrast contours(OC) elicits larger TAEs than does adaptation to SCs,suggesting that they more effectively activate neurons inV1/V2 during prolonged viewing. This could be due tothe integration of orientation contrast signals in a feed-forward process, which provides a more robust contourorientation signal than can be compensated for by feed-back from higher cortical areas. However, the encoding oforientation for both contour types may rely on feedback tosome extent, as they adapted an LC as strongly as did anLC itself. This demonstrates shared sites of orientationencoding in V1/V2 between first- and second-order orien-tation signals. Hence, while SCs do not appear to be asrobust an orientation signal as texture contours containingan additional orientation contrast cue, their orientationis at least encoded by the same population of low-levelorientation-tuned neurons.

Acknowledgment

This research was supported by an EPSRC studentshipand EPSRC grand no. GR/N26296/01. The authors wishto thank M. Cox for comments on an earlier version of themanuscript.

Commercial relationships: none.Corresponding author: Dr. Sarah J. Hawley.Email: harrison.sarahj@gmail.com.Address: School of Psychology, Adeilad Brigantia,University of Wales Bangor, Gwynedd, LL57 2AS, UK.

References

Baumann, R., van der Zwan, R., & Peterhans, E. (1997).Figure-ground segregation at contours: A neural mech-anism in the visual cortex of the alert monkey.EuropeanJournal of Neuroscience, 9(6), 1290Y1303. [PubMed]

Baylis, G. C., & Driver, J. (2001). Shape-coding in ITcells generalizes over contrast and mirror reversal,but not figure-ground reversal. Nature Neuroscience,4(9), 937Y942. [PubMed] [Article]

Beck, J. (1966). Effect of orientation and of shapesimilarity on perceptual grouping. Perception &Psychophysics, 1, 300Y302.

Bergen, J. R. (1991). Theories of visual texture percep-tion. In D. Regan (Ed.), Spatial Vision (Vol. 10, pp.114Y134). London: Macmillan.

Bergen, J. R., & Landy, M. S. (1991). Computational mod-eling of visual texture segregation. In M. S. Landy &M. J. Anthony (Eds.), Computational modelling of vi-sual processing. Cambridge, Mass London: MIT Press.

Berkley, M. A., Debruyn, B., & Orban, G. (1994).Illusory, motion, and luminance-defined contoursinteract in the human visual system. Vision Research,34(2), 209Y216. [PubMed]

Blakemore, C., Carpenter, R. H., & Georgeson, M. A.(1970). Lateral inhibition between orientation detec-tors in the human visual system. Nature, 228(5266),37Y39. [PubMed]

Bockisch, C. J. (1999). Orientation discrimination and tiltaftereffects with luminance and illusory contours.Vision Research, 39(4), 765Y776. [PubMed]

Bradley, D. R., & Dumais, S. T. (1984). The effects ofillumination level and retinal size on the depth strat-ification of subjective contour figures. Perception,13(2), 155Y164. [PubMed]

Carpenter, R. H., & Blakemore, C. (1973). Interactionsbetween orientations in human vision. ExperimentalBrain Research, 18(3), 287Y303. [PubMed]

Journal of Vision (2006) 6, 37–52 Hawley & Keeble 49

Clifford, C. W., Pearson, J., Forte, J. D., & Spehar, B.(2003). Colour and luminance selectivity of spatialand temporal interactions in orientation perception.Vision Research, 43(27), 2885Y2893. [PubMed]

Finney, D. J. (1971). Probit analysis. London: CambridgeUniversity Press.

Frisby, J. P., & Clatworthy, J. L. (1975). Illusory con-tours: Curious cases of simultaneous brightness con-trast? Perception, 4, 349Y357.

Georgeson, M. A., & Schofield, A. J. (2002). Shading andtexture: Separate information channels with a com-mon adaptation mechanism? Spatial Vision, 16(1),59Y76. [PubMed]

Gibson, J. J., & Radner, M. (1937). Adaptation, aftereffectand contrast in the perception of tilted lines: I.quantitave studies. Journal of Experimental Psychol-ogy, 20, 453Y467.

Graham, N. (1994). Non-linearities in texture segregation.Ciba Foundation Symposium, 184, 309Y322; discus-sion 323Y338. [PubMed]

Grill-Spector, K., Kourtzi, Z., & Kanwisher, N. (2001).The lateral occipital complex and its role in objectrecognition. Vision Research, 41(10Y11), 1409Y1422.[PubMed]

Gulyas, B., Cowey, A., Heywood, C. A., Popplewell, D.,& Roland, P. E. (1998). Visual form discriminationfrom texture cues: A PET study. Human BrainMapping, 6(3), 115Y127. [PubMed]

Harris, J. P., & Calvert, J. E. (1989). Contrast, spatialfrequency and test duration effects on the tilt after-effect: Implications for underlying mechanisms.Vision Research, 29(1), 129Y135. [PubMed]

Ito, M., & Gilbert, C. D. (1999). Attention modulatescontextual influences in the primary visual cortex ofalert monkeys. Neuron, 22(3), 593Y604. [PubMed]

Johnstone, S., & Wenderoth, P. (1989). Pattern orienta-tion, not motion, determines the two-dimensional tiltillusion. Perception, 18(6), 729Y737. [PubMed]

Joseph, J. S., & Optican, L. M. (1996). Involuntary at-tentional shifts due to orientation differences. Percep-tion & Psychophysics, 58(5), 651Y665. [PubMed]

Julesz, B. (1981). Textons, the elements of texture per-ception, and their interactions. Nature, 290(5802),91Y97. [PubMed]

Julesz, B., & Bergen, J. R. (1983). Textons, the funda-mental elements in preattentive vision and percep-tion of textures. Bell System Technical Journal, 62(6),1619Y1645.

Kanizsa, G. (1976). Subjective contours. Scientific Amer-ican, 234(4), 48Y52. [PubMed]

Kourtzi, Z., Tolias, A. S., Altmann, C. F., Augath, M., &Logothetis, N. K. (2003). Integration of local featuresinto global shapes: Monkey and human FMRIstudies. Neuron, 37(2), 333Y346. [PubMed]

Kwan, L., & Regan, D. (1998). Orientation-tuned spatialfilters for texture-defined form. Vision Research,38(24), 3849Y3855. [PubMed]

Lamme, V. A., & Roelfsema, P. R. (2000). The distinctmodes of vision offered by feedforward and recurrentprocessing. Trends in Neuroscience, 23(11), 571Y579.[PubMed]

Lamme, V. A., Super, H., & Spekreijse, H. (1998).Feedforward, horizontal, and feedback processing inthe visual cortex. Current Opinion in Neurobiology,8(4), 529Y535. [PubMed]

Lamme, V. A., Zipser, K., & Spekreijse, H. (1998).Figure-ground activity in primary visual cortex is sup-pressed by anesthesia. Proceedings of the NationalAcademy of Sciences of the United States of America,95(6), 3263Y3268. [PubMed] [Article]

Landy, M. S., & Bergen, J. R. (1991). Texture segregationand orientation gradient. Vision Research, 31(4),679Y691. [PubMed]

Lee, T. S., & Mumford, D. (2003). Hierarchical Bayesianinference in the visual cortex. Journal of the OpticalSociety of America. A, Optics, Image Science, andVision, 20(7), 1434Y1448. [PubMed]

Lee, T. S., Mumford, D., Romero, R., & Lamme, V. A.(1998). The role of the primary visual cortex inhigher level vision. Vision Research, 38(15Y16),2429Y2454. [PubMed]

Lee, T. S., & Nguyen, M. (2001). Dynamics of subjectivecontour formation in the early visual cortex.Proceedingsof the National Academy of Sciences of the United Statesof America, 98(4), 1907Y1911. [PubMed] [Article]

Lee, T. S., Yang, C. F., Romero, R. D., & Mumford, D.(2002). Neural activity in early visual cortex reflectsbehavioral experience and higher-order perceptual sa-liency. Nature Neuroscience, 5(6), 589Y597. [PubMed][Article]

Leventhal, A. G., Wang, Y., Schmolesky, M. T., & Zhou,Y. (1998). Neural correlates of boundary perception.Visual Neuroscience, 15(6), 1107Y1118. [PubMed]

Li, Z. (2000). Pre-attentive segmentation in the primaryvisual cortex. Spatial Vision, 13(1), 25Y50. [PubMed]

Magnussen, S., & Kurtenbach, W. (1980a). Adapting totwo orientations: Disinhibition in a visual aftereffect.Science, 207(4433), 908Y909. [PubMed]

Magnussen, S., & Kurtenbach, W. (1980b). Linearsummation of tilt illusion and tilt aftereffect. VisionResearch, 20(1), 39Y42. [PubMed]

Journal of Vision (2006) 6, 37–52 Hawley & Keeble 50

Malik, J., & Perona, P. (1990). Preattentive texturediscrimination with early vision mechanisms. Journalof the Optical Society of America A, 7(5), 923Y932.[PubMed]

Marcus, D. S., & Van Essen, D. C. (2002). Scene segmen-tation and attention in primate cortical areas V1 andV2. Journal of Neurophysiology, 88(5), 2648Y2658.[PubMed] [Article]

Montaser-Kouhsari, L., & Rajimehr, R. (2004). Attentionalmodulation of adaptation to illusory lines. Journalof Vision, 4(6), 434Y444, http://www.journalofvision.org/4/6/3/, doi:10.1167/4.6.3. [PubMed] [Article]

Morrone, M. C., Burr, D. C., & Maffei, L. (1982).Functional implications of cross-orientation inhibitionof cortical visual cells: I. Neurophysiological evidence.Proceedings of the Royal Society of London. Series B,Biological Sciences, 216(1204), 335Y354. [PubMed]

Motoyoshi, I., & Nishida, S. (2001). Visual responsesaturation to orientation contrast in the perceptionof texture boundary. Journal of the Optical Societyof America. A, Optics, Image Sciences, and Vision,18(9), 2209Y2219. [PubMed]

Nothdurft, H. C. (1985). Orientation sensitivity and texturesegmentation in patterns with different line orientation.Vision Research, 25(4), 551Y560. [PubMed]

Nothdurft, H. C. (2002). Attention shifts to salient targets.Vision Research, 42(10), 1287Y1306. [PubMed]

Nothdurft, H. C., & Li, C. Y. (1985). Texture discrim-ination: Representation of orientation and luminancedifferences in cells of the cat striate cortex. VisionResearch, 25(1), 99Y113. [PubMed]

Paradiso, M. A., Shimojo, S., & Nakayama, K. (1989).Subjective contours, tilt aftereffects, and visual cor-tical organization. Vision Research, 29(9), 1205Y1213.[PubMed]

Parker, D. M. (1972). Contrast and size variables and thetilt after-effect. Quarterly Journal of ExperimentalPsychology, 24(1), 1Y7. [PubMed]

Peterhans, E., & von der Heydt, R. (1989). Mechanismsof contour perception in monkey visual cortex. II.Contours bridging gaps. Journal of Neuroscience,9(5), 1749Y1763. [PubMed] [Article]

Poom, L. (2000). Inter-attribute tilt effects and orientationanalysis in the visual brain. Vision Research, 40(20),2711Y2722. [PubMed]

Popple, A. V. (2003). Context effects on texture border local-ization bias. Vision Research, 43(7), 739Y743. [PubMed]

Press,W. H. (1992).Numerical recipes in C: The art of scien-tific computing. Cambridge: Cambridge University Press.

Prins, N., & Mussap, A. J. (2001). Adaptation reveals aneural code for the visual location of orientationchange. Perception, 30(6), 669Y680. [PubMed]

Pritchard, W. S., & Warm, J. S. (1983). Attentional pro-cessing and the subjective contour illusion. Journal ofExperimental Psychology: General, 112(2), 145Y175.[PubMed]

Ramsden, B. M., Hung, C. P., & Roe, A. W. (2001). Realand illusory contour processing in area V1 of theprimate: A cortical balancing act. Cerebral Cortex,11(7), 648Y665. [PubMed] [Article]

Redies, C., Crook, J. M., & Creutzfeldt, O. D. (1986).Neuronal responses to borders with and withoutluminance gradients in cat visual cortex and dorsallateral geniculate nucleus. Experimental BrainResearch, 61(3), 469Y481. [PubMed]

Regan, D. (1995). Orientation discrimination for barsdefined by orientation texture. Perception, 24(10),1131Y1138. [PubMed]

Sheth, B. R., Sharma, J., Rao, S. C., & Sur, M. (1996).Orientation maps of subjective contours in visualcortex. Science, 274(5295), 2110Y2115. [PubMed][Article]

Skillen, J., Whitaker, D., Popple, A. V., & McGraw, P. V.(2002). The importance of spatial scale in determin-ing illusions of orientation. Vision Research, 42(21),2447Y2455. [PubMed]

Skottun, B. C., Johnsen, T., & Magnussen, S. (1981). Tiltaftereffect with small adapting angles. Perception &Psychophysics, 30(2), 199Y200. [PubMed]

Smith, S., Wenderoth, P., & van der Zwan, R. (2001).Orientation processing mechanisms revealed by theplaid tilt illusion. Vision Research, 41(4), 483Y494.[PubMed]

Snowden, R. J. (1994). Adaptability of the visual systemis inversely related to its sensitivity. Journal of theOptical Society of America A, 11(1), 25Y32. [PubMed]

Soriano, M., Spillmann, L., & Bach, M. (1996). Theabutting grating illusion. Vision Research, 36(1),109Y116. [PubMed]

Super, H., Spekreijse, H., & Lamme, V. A. (2001). Twodistinct modes of sensory processing observed inmonkey primary visual cortex (V1). Nature Neuro-science, 4(3), 304Y310. [PubMed] [Article]

Tanaka, K. (2001). Late responses and perceptual aware-ness. Nature Neuroscience, 4(3), 225Y226. [PubMed][Article]

Treisman, A. (1985). Preattentive processing in vision.Computer Vision, Graphics, and Image Processing,31, 156Y177.

van der Zwan, R., & Wenderoth, P. (1994). Psychophys-ical evidence for area V2 involvement in the re-duction of subjective contour tilt aftereffects bybinocular rivalry. Vision Neuroscience, 11(4),823Y830. [PubMed]

Journal of Vision (2006) 6, 37–52 Hawley & Keeble 51

van der Zwan, R., & Wenderoth, P. (1995). Mechanismsof purely subjective contour tilt aftereffects. VisionResearch, 35(18), 2547Y2557. [PubMed]

Vogels, R., & Orban, G. A. (1987). Illusory contourorientation discrimination. Vision Research, 27(3),453Y467. [PubMed]

von der Heydt, R., & Peterhans, E. (1989). Mechanismsof contour perception in monkey visual cortex. I.Lines of pattern discontinuity. Journal of Neuro-science, 9(5), 1731Y1748. [PubMed] [Article]

von der Heydt, R., Peterhans, E., & Baumgartner, G.(1984). Illusory contours and cortical neuron re-sponses. Science, 224(4654), 1260Y1262. [PubMed]

Wenderoth, P., & Curthoys, I. (1974). On the non-additivity of visual tilt illusions. Quarterly Journal ofExperimental Psychology, 26(4), 549Y555. [PubMed]

Wenderoth, P., & Johnstone, S. (1987). Possible neuralsubstrates for orientation analysis and perception.Perception, 16(6), 693Y709. [PubMed]

Wenderoth, P., & Johnstone, S. (1988). The differentmechanisms of the direct and indirect tilt illusions.Vision Research, 28(2), 301Y312. [PubMed]

Wenderoth, P., Johnstone, S., & van der Zwan, R. (1989).Two-dimensional tilt illusions induced by orthogonalplaid patterns: Effects of plaid motion, orientation,spatial separation, and spatial frequency. Perception,18(1), 25Y38. [PubMed]

Wenderoth, P., van der Zwan, R., & Johnstone, S. (1989).Orientation illusions induced by briefly flashedplaids. Perception, 18(6), 715Y728. [PubMed]

Wenderoth, P., van der Zwan, R., & Williams, M. (1993).Direct evidence for competition between local andglobal mechanisms of two-dimensional orientationillusions. Perception, 22(3), 273Y286. [PubMed]

Westheimer, G., & Li, W. (1996). Classifying illusory con-tours by means of orientation discrimination. Journal ofNeurophysiology, 75(2), 523Y537. [PubMed]

Westheimer, G., & Li, W. (1997). Classifying illusorycontours: Edges defined by Bpacman[ and monoculartokens. Journal of Neurophysiology, 77(2), 731Y736.[PubMed] [Article]

Whitaker, D., McGraw, P. V., & Levi, D. M. (1997). Theinfluence of adaptation on perceived visual location.Vision Research, 37(16), 2207Y2216. [PubMed]

Wolfe, J. M. (1984). Short test flashes produce large tiltaftereffects. Vision Research, 24(12), 1959Y1964.[PubMed]

Wolfe, J. M., & O’Connell, K. M. (1986). Fatigue andstructural change: Two consequences of visualpattern adaptation. Investigative Ophthalmology &Visual Science, 27(4), 538Y543. [PubMed]

Wolfson, S. S., & Landy, M. S. (1999). Long rangeinteractions between oriented texture elements.Vision Research, 39(5), 933Y945. [PubMed]

Journal of Vision (2006) 6, 37–52 Hawley & Keeble 52