Long range interactions between oriented texture elements.

Long range interactions between texture elements (short, oriented line segments) were examined. Specifically, we studied the influence of a background array of texture elements on the detectability of a target element (separated from the background by an intermediate textured region) using textures like those of Caputo (Vis. Res. 1996, 36, 2815-2826). We found that, in general, when the background elements were oriented orthogonally to the target element, detection of the target element was better than when the background elements had the same orientation as the target element. We discuss these interactions in terms of inhibitory and excitatory connections between orientation and spatial frequency selective linear filters (e.g. filters which mimic V1 simple cells) which would respond to the individual texture elements.     

Predicting the readability of transparent text

Will a simple global masking model based on image detection be successful at predicting the readability of transparent text? Text readability was measured for two types of transparent text: additive (as occurs in head-up displays) and multiplicative (which occurs in see-through liquid crystal display virtual reality displays). Text contrast and background texture were manipulated. Data from two previous experiments were also included (one using very low contrasts on plain backgrounds, and the other using higher-contrast opaque text on both plain and textured backgrounds). All variables influenced readability in at least an interactive manner. When there were background textures, the global masking index (that combines text contrast and background root mean square contrast) was a good predictor of search times (r = 0.89). When the masking was adjusted to include the text pixels as well as the background pixels in computations of mean luminance and contrast variability, predictability improved further (r = 0.91).     

Properties of second-order spatial frequency channels

The segregation of texture patterns may be carried out by a set of linear spatial filters (to enhance one of the constituent textures), a nonlinearity (to convert the higher contrast of response to that constituent to a higher mean response), and finally subsequent ("second-order") linear spatial filters (to provide a strong response to the texture-defined edge itself). In this paper, the properties of such second-order filters are characterized. Observers were required to detect or discriminate textures that were modulated between predominantly horizontally oriented and predominantly vertically oriented noise patterns. Spatial summation for these patterns reached asymptote for a stimulus size of 15 x 15 deg. Modulation contrast sensitivity was nearly flat over a five-octave range of spatial frequency, but was bandpass when stated as efficiency (relative to an idealized observer confronted with the same task). Increment threshold showed the improved performance with a sub-threshold pedestal seen in the "dipper effect", but the typical Weber's law behavior at higher pedestal contrasts was not observed at the highest pedestal modulation contrasts achievable with our stimuli. Sub-threshold summation experiments indicate that second-order filters have a moderate bandwidth.     

A comparison of the dynamics of simple (Fourier) and complex (non-Fourier) mechanisms in texture segregation.

Models of texture segregation frequently feature two processing mechanisms: simple, linear channels (1st-order, Fourier mechanisms) and complex channels (2nd-order, non-Fourier mechanisms). Using texture patterns designed to segregate primarily as a result of activity in one set of channels or the other, we employed the method of cued response to obtain speed-accuracy tradeoff (SAT) functions measuring the time course of texture segregation processing in simple and complex channels. Here, both simple-channel and complex-channel patterns are composed of Gabor-patch texture elements, thus equating input to simple channels and the first stage of complex channels. Subjects were required to identify the orientation of a rectangular texture-region embedded in a background field of a different texture. SAT functions were obtained by requiring subjects to respond within 200 ms after an auditory cue. We found that: (1) when segregation depended primarily on activity in simple channels, performance was faster and better than when it depended primarily on complex channels; (2) in contrast to a previous study (Sutter, A., & Graham, N. (1995). Investigating simple and complex mechanisms in texture segragation using the speed-accuracy tradeoff method. Vision Research, 35, 2825-2843), simple-channel (Fourier) patterns composed of two textured regions were just as easily segregated as simple-channel patterns in which one of the regions was blank instead of textured; (3) performance with complex-channel patterns composed of diagonally oriented Gabor-patches was considerably worse than performance with complex-channel patterns composed of vertically and/or horizontally oriented Gabor-patches; (4) among simple-channel patterns containing only one region of texture (background-only or rectangle-only), there were minimal differences in performance; and (5) as in previous experiments, there were large individual differences in the segregation of complex-channel (non-Fourier) patterns. All of the above results can be explained within the framework of the simple- and complex-channels model of texture segregation.     

Segregation of mesh-derived textures evaluated by resistance to added disorder.

In the "figure detection task" the strength of segregation for a particular texture pair was estimated by the threshold amount of added disorder that prevented segregation of a textured figure from a textured ground. Disorder was either jitter in the orientation of the texture elements, or jitter in their xy positions, or a mixture of the two. Other procedures included lowpass filtering, and a task requiring discrimination between textured figures of different shapes. Orientation cues are weakly or inconsistently used for segregating mesh textures. The low spatial harmonics are very important. A new finding is that orientation and position jitter thresholds for a set of figure/ground texture patterns are often proportional. In a mixture the one disorder can be exchanged for the other.     

Texture segregation on the basis of contrast polarity of odd-symmetric filters

This is the first study to demonstrate the selectivity of learning for contrast polarity. The finding is the main result of an investigation into the existence of central and peripheral vision mechanisms selective for contrast polarity within the texture-segregation process, using the perceptual learning paradigm in a detection task. Energy models (Malik & Perona, 1990) exclude segregation of textures composed of elements of odd-symmetric luminance profile by contrast polarity differences. Here the target was a Gabor patch (0.8 deg) of 1 cyc/deg in sine phase (odd-symmetry) embedded in a background of mirror-image elements. Our results showed that, in fovea, segregation on the basis of contrast polarity was above threshold from the first session. After learning, the target popped-out in both central and peripheral vision for durations over 10 ms. Our major result is that learning is selective for contrast polarity; it is also selective for orientation and position, all characteristics distinctive of early processing. Since the learning effects were obtained with texture composed of odd-symmetric mirror-image elements, they indicate that the output from odd-symmetric filters was not excluded or inhibited in texture segmentation, but instead played an active role. Our data support models of texture segmentation, in which detection of texture gradient is achieved on the basis of early cortical process, before the non-linear transformation of their output.     

Multiple channels for horizontal, but only one for vertical corrugations? A new look at the stereo anisotropy

Stereo vision displays a well-known anisotropy: disparity-defined slant is easier to detect for rotations about a horizontal axis than about a vertical axis, and low-frequency sinusoidal depth corrugations are easier to detect when the corrugations are horizontal than when they are vertical. Here, we determined disparity thresholds for vertically and horizontally oriented depth corrugations with both sinusoidal and square-wave profiles. We found that the orientation anisotropy for square waves is much weaker than for sine waves and is almost independent of frequency. This weaker anisotropy for square waves can be explained by considering the Fourier harmonics present in the stimulus. Using linear models imported from the luminance and texture perception domain, the disparity thresholds for square waves can be very well predicted from those for sine waves, for both horizontally and vertically oriented corrugations. For horizontally oriented corrugations, models based on the root mean square of the output of a single linear channel or the output of multiple linear channels worked equally well. This is consistent with previous evidence suggesting that stereo vision has multiple channels tuned to different spatial frequencies of horizontally oriented disparity modulations. However, for vertically oriented corrugations, only the root mean squared output of a single linear channel explained the data. We suggest that the stereo anisotropy may arise because the stereo system possesses multiple spatial frequency channels for detecting horizontally oriented modulations in horizontal disparity, but only one for vertically oriented modulations.     

Backward masking and the central performance drop

Kehrer [Spatial Vision 2 (1987) 247] found that texture discrimination performance sometimes peaks in the parafovea rather than at the fovea, and he referred to this phenomenon as the central performance drop (CPD). Kehrer used a backward mask to limit performance and Morikawa [Vision Res. 40 (2000) 3517] argued that in some cases the temporal aspects of the backward mask may be critical to the emergence of the CPD. In one experiment Morikawa showed that the CPD does not emerge when a simultaneous noise-mask (different from the mask used by Kehrer) is used to limit performance. In another experiment Morikawa showed that unmasked texture displays comprising short lines do not elicit the CPD. In both cases, changes in the temporal aspects of the texture displays were accompanied by changes in the spatial structure of the mask or stimulus. For the spatio-temporal theory of the CPD to be sustained one would have to show that noise masks elicit a CPD when used as backward masks and that the short-line textures elicit a CPD when followed by backward masks. Our evidence provides little if any support for either of these predictions. Furthermore, an analysis of a simple filter-rectify-filter model of texture segmentation shows that a greatly attenuated CPD is to be expected when a noise mask is used as a source of spatial noise.     

Automatic texture segmentation in early vision: evidence from priming experiments

Texture segmentation is usually regarded as a fast, early, automatic, preattentive process. Nevertheless, naive participants in texture segmentation tasks are usually not able to perform the task explicitly when the textures are presented rather briefly (49 ms) and subsequently masked. In two experiments it was investigated whether texture stimuli were, nevertheless, automatically segmented under these conditions. By using a priming paradigm, the processing of the texture stimuli was measured indirectly via their influence on a subsequently presented imperative stimulus. Priming effects were found for experienced and naive participants, although novices could not respond overtly to the textures in a subsequent forced-choice task. Although textures influenced subsequent stimulus processing, an analysis of the simultaneously recorded lateralized readiness potential (LRP) showed that they did not cause automatic response activation. The existence of priming effects of textures without participants' ability to overtly respond to them can be regarded as evidence for the automatic segmentation of texture stimuli.     

Perception of three-dimensional shape from texture is based on patterns of oriented energy

This paper presents empirical support for a new observer model of inferring three-dimensional shape from monocular texture cues. By measuring observers' abilities to estimate the relative three-dimensional curvature along a textured surface from two-dimensional projected images, and concurrently examining the local spectral changes occurring in the projected image for various texture patterns, we have found that correlated changes in oriented energy along lines corresponding to the lines of maximum and minimum curvature of the surface are crucial for conveying the three-dimensional shape of the surface. Energy along these lines of maximum and minimum curvature can be used to compute the orientation of local surface patches. Texture patterns consisting of simple and complex sinusoidal gratings and plaids, and filtered noise were drawn onto a surface that was corrugated sinusoidally in depth about the horizontal axis and projected in perspective onto an image plane. The perceived relative surface curvature was reconstructed from measurements of local ordinal depth around a central fixation point at 12 different phases of the corrugation. Our results show that: (1) it is neither necessary nor sufficient to identify individual texture elements or texture gradients in order to extract the shape of the surface; (2) one-dimensional frequency modulation is insufficient for conveying complex three-dimensional shape. (3) Veridical ordinal depth is seen only when the projected pattern contains changes in oriented energy along lines corresponding to projected lines of maximum curvature of the surface. (4) For a surface corrugated in depth about the horizontal axis, this pattern of oriented energy arises from energy along the vertical direction in the global Fourier transform of the pre-corrugated pattern. (5) Local orientation changes across lines of minimum curvature can be also critical for conveying shape. (6) These correlated orientation changes along lines of maximum and minimum curvature are entirely lost in parallel projection. Hence texture is a useful cue for shape if the image is a perspective projection. (7) Only some natural textures will provide sufficient monocular cues to support veridical shape inferences, and this can be predicted from their global Fourier transforms.     

Saliency and context play a role in infants' texture segmentation

We investigated whether young infants orient reliably towards more salient vs. less salient objects in a visual scene. Subjects were tested with stimuli presented on textured fields, one side showing a target stimulus (a 'more salient' or 'less salient' texture patch) and the other a background stimulus. Infants typically preferred the more salient, but not the less salient target. Their behaviour depended on the configuration of the background stimulus. In contrast, 3-4 year-old children always showed a preference for the target stimulus, regardless of the configuration of the background. We conclude that both saliency of a target stimulus and its context play a role in early texture segmentation.     

Figure-ground interaction in the human visual cortex

Discontinuities in feature maps serve as important cues for the location of object boundaries. Here we used multi-input nonlinear analysis methods and EEG source imaging to assess the role of several different boundary cues in visual scene segmentation. Synthetic figure/ground displays portraying a circular figure region were defined solely by differences in the temporal frequency of the figure and background regions in the limiting case and by the addition of orientation or relative alignment cues in other cases. The use of distinct temporal frequencies made it possible to separately record responses arising from each region and to characterize the nature of nonlinear interactions between the two regions as measured in a set of retinotopically and functionally defined cortical areas. Figure/background interactions were prominent in retinotopic areas, and in an extra-striate region lying dorsal and anterior to area MT+. Figure/background interaction was greatly diminished by the elimination of orientation cues, the introduction of small gaps between the two regions, or by the presence of a constant second-order border between regions. Nonlinear figure/background interactions therefore carry spatially precise, time-locked information about the continuity/discontinuity of oriented texture fields. This information is widely distributed throughout occipital areas, including areas that do not display strong retinotopy.     

Higher order texture statistics impair contrast boundary segmentation

Texture boundary segmentation is conventionally thought to be mediated by global differences in Fourier energy, i.e., low-order texture statistics. Here, we have examined the importance of higher order statistical structure of textures in a simple second-order segmentation task. We measured modulation depth thresholds for contrast boundaries imposed on texture samples extracted from natural scene photographs, using forced-choice judgments of boundary orientation (left vs. right oblique). We compared segmentation thresholds for contrast boundaries whose constituent textures were either intact or phase scrambled. In the intact condition, all the texture statistics were preserved, while in the phase-scrambled condition the higher order statistics of the same texture were randomized, but the lower order statistics were unchanged. We found that (1) contrast boundary segmentation is impaired by the presence of higher order statistics; (2) every texture shows impairment but some substantially more than others; and (3) our findings are not related to scrambling-induced changes in detectability. The magnitude of phase-scrambling effect for individual textures was uncorrelated with variations in their amplitude spectra, but instead we suggest that it might be related to differences in local edge structure or sparseness.     

Feature-specific electrophysiological correlates of texture segregation

Discrimination between a figure and its surround is an important first step of pattern recognition. This discrimination usually relies, as a first step, on the detection of borders between a figure and its surround, for example based on spatial gradients in luminance, colour, or texture. There is evidence that neurones in the visual cortex are specifically activated by segregation between textures, but the relation between segregation based on different types of features such as colour, luminance, and motion is unclear. Evoked EEG potentials specific to texture segregation were investigated in 17 observers in two separate experiments and by means of functional magnetic resonance imaging in a separate study (Fahle et al., in preparation). Differences in either luminance, colour, line orientation, motion, or stereoscopic depth defined a checkerboard pattern. Patterns defined by each of these features elicited segregation-specific potentials. In contrast to earlier reports (Vision Research 37 (1997) 1409), however, we find pronounced differences between the segregation-specific potentials evoked through different features, especially regarding their peak latencies. The topographical distribution of the activity evoked reveals different polarities and partly specific locations for different stimulus features, indicating the existence of different processors for texture segregation based on different features.     

Contrasting the processes of texture segmentation and discrimination with static and phase-reversing stimuli

Regions of visual texture can be automatically segregated from one another when they abut but also discriminated from one another if they are separated in space or time. A difference in mean orientation between two textures serves to facilitate their segmentation, whereas a difference in orientation variance does not. The present study further supports this notion, by replicating the findings of Wolfson and Landy (1998) in showing that judgments (odd-one-out) made for textures that differ in mean orientation were more accurate (and more rapid) when the textures were abutting than when separated, whereas judgments of variance were made no more accurately for abutting relative to separated textures. Interestingly, however, responses were overall faster for textures differing in variance when they were separated compared to when they were abutting. This is perhaps due to the clear separation boundary, which serves to delineate the regions on which to perform some regional estimation of orientation variance. A second experiment highlights the phase-insensitivity of texture segmentation, in that locating a texture edge (defined by a difference in mean orientation) in high frequency orientation-reversing stimuli can be performed at much higher frequencies than the discrimination of the same regions but with the texture contour masked. Textures that differed in variance did not exhibit this effect. A final experiment demonstrates that the phase-insensitive perception of texture borders improves with eccentric viewing relative to the fovea, whereas perception of the texture regions does not. Together, these experiments show dissociations between edge- and region-based texture analysis mechanisms and suggest a fast, sign-invariant contour extraction system mediating texture segmentation, which may be closely linked to the magnocellular subdivision of visual processing.     

Texture segregation in the cat: a parametric study.

We have investigated how different texture parameters affect texture segregation in the cat, and which strategies cats use to solve the segregation task. Five cats were presented with stimuli consisting of two adjacent panels. One side contained a square area of a particular texture embedded in a different background texture; the other side was filled with only the background texture. The animal's task was to detect at which side the texture difference was presented. Sensitivity for the texture difference was assessed by making one aspect of the texture (in most instances the size of the texture elements) dependent upon performance by means of a staircase procedure. Among the most prominent parametric effects are those of density and element position randomization. In general, segregation was optimal at intermediate densities and deteriorated at larger and smaller densities. Element position randomization caused a slight but systematic decrease in segregation performance. Furthermore, we found texture elements at the border between different textures to be of primary importance for segregation. Which strategy the animals used for solving the segregation task depended upon the presence of random figure/background reversals in subsequent stimulus presentations during training. The animals learned to detect texture differences if these reversals were present, and without reversals, they learned to identify the particular texture in the target square. Interestingly, parameter dependencies of segregation did not depend upon the detection strategy used. We have speculated that the two different strategies used by the cats to solve the segregation tasks are related to different hierarchical levels of texture segregation which can be traced back to different stages of texture processing in human models of segregation performance.     

A visual evoked potential correlate of global figure-ground segmentation.

Human observers discriminated the global orientation of a texture-defined figure which segregated from a texture surround. Global figure discriminability was manipulated through within-figure collinearity, figure-surround interaction, and figure connectedness, while the local orientation contrast at edges between figure and surround was kept constant throughout all the experiments. Visual evoked potentials (VEPs) were recorded during onset-offset stimulation in which the figure cyclically appeared and disappeared from a uniform texture background. A difference component was obtained by subtraction of offset-from onset-VEP. Two negative peaks of the difference component are found with latencies around 140-160 and 200-260 ms, respectively. Enhanced discriminability of the global figure reduced (11-25 ms) the latency of the second peak, hence indicating that the 200-260 ms component was produced by global figure-ground segmentation.     

Effect of context and efference copy on visual straight ahead

Bias in efferent commands to the eye changes the apparent straight ahead direction in an unstructured visual field, but has little effect in a normal visual environment. Naive subjects set a visible marker to appear straight ahead under monocular viewing conditions and while pressing on the viewing eye. Three background conditions were used: a naturalistic landscape photograph, a blank field, and a repeating checkerboard texture that provides strong contours but no information about visual direction. Effect of eyepress on straight-ahead judgments was small but significant with the landscape background, and larger with the blank field; the checkerboard texture yielded a bias halfway between the magnitudes of bias in the other two conditions. A visual capture theory predicts that the textured field should work like a blank one, while an oculomotor theory predicts that it should work like a natural one. Interpreted in this context, the results show the two theories to be about equally important in judging straight ahead. A second experiment with experienced observers and moving backgrounds gave the same result.     

Spatial-bisection acuity in infantile nystagmus

This study measured spatial bisection acuity for horizontally and vertically separated line targets in five observers with infantile nystagmus syndrome (INS) and no obvious associated sensory abnormalities, and in two normal observers during comparable horizontal retinal image motion. For small spatial separations between the line targets, bisection acuity for both horizontally and vertically separated lines is worse in the observers with IN than normal observers. In four of the five observers with IN, bisection acuity for small target separations is poorer for horizontally compared to vertically separated lines. Because the motion smear generated by the retinal image motion during IN would be expected to influence horizontally separated targets, the degradation of bisection acuity for both vertical and horizontally separated lines indicates that a sensory neural deficit contributes to impaired visual functioning in observers with idiopathic IN.     

Some observations on the effects of slant and texture type on slant-from-texture

We measure the performance of five subjects in a two-alternative-forced-choice slant-discrimination task for differently textured planes. As textures we used uniform lattices, randomly displaced lattices, circles (polka dots), Voronoi tessellations, plaids, 1/f noise, "coherent" noise and a leopard skin-like texture. Our results show: (1) Improving performance with larger slants for all textures, (2) and some cases of "non-symmetrical" performance around a particular orientation. (3) For orientations sufficiently slanted, the different textures do not elicit major differences in performance, (4) while for orientations closer to the vertical plane there are marked differences among them. (5) These differences allow a rank-order of textures to be formed according to their "helpfulness"--that is, how easy the discrimination task is when a particular texture is mapped on the plane. Polka dots tend to allow the best slant discrimination performance, noise patterns the worst. Two additional experiments were conducted to test the generality of the obtained rank-order. First, the tilt of the planes was rotated by 90 degrees. Second, the task was changed to a slant report task via probe adjustment. The results of both control experiments confirmed the texture rank-order previously obtained. We then test a number of spatial-frequency-based slant-from-texture models and discuss their shortcomings in explaining our rank-order. Finally, we comment on the importance of these results for depth-perception research in general, and in particular the implications our results have for studies of cue combination (sensor fusion) using texture as one of the cues involved.