A neural network model for texture discrimination

A model of texture discrimination in visual cortex was built using a feedforward network with lateral interactions among relatively realistic spiking neural elements. The elements have various membrane currents, equilibrium potentials and time constants, with action potentials and synapses. The model is derived from the modified programs of MacGregor (1987). Gabor-like filters are applied to overlapping regions in the original image; the neural network with lateral excitatory and inhibitory interactions then compares and adjusts the Gabor amplitudes in order to produce the actual texture discrimination. Finally, a combination layer selects and groups various representations in the output of the network to form the final transformed image material. We show that both texture segmentation and detection of texture boundaries can be represented in the firing activity of such a network for a wide variety of synthetic to natural images. Performance details depend most strongly on the global balance of strengths of the excitatory and inhibitory lateral interconnections. The spatial distribution of lateral connective strengths has relatively little effect. Detailed temporal firing activities of single elements in the lateral connected network were examined under various stimulus conditions. Results show (as in area 17 of cortex) that a single element's response to image features local to its receptive field can be altered by changes in the global context.     

Some psychophysical determinants of discrete Moiré patterns

In a series of experiments we have investigated the perception of Moiré patterns as a function of spatial density, rotation and temporal display parameters. Results indicate that the local correlation extraction process involved in the perception of these patterns is not feature specific, yet is driven by excitatory (correlated) and inhibitory (uncorrelated) information under a form of spatial summation. These results are comparable with recent results on texture discrimination where texture interpoint distance distributions (dipole statistics) have also been discovered to have excitatory and inhibitory components.     

Texture discrimination asymmetries across the visual field

Texture discrimination is sometimes asymmetrical; texture A embedded in texture B is more easily detected than texture B embedded in texture A. Furthermore, texture discrimination often improves as the disparate texture is moved into the periphery; this has been referred to as the central performance drop (CPD). The interaction of these interesting and counter-intuitive aspects of texture discrimination has received very little attention. Using four stimulus pattern pairs that were previously shown to elicit asymmetrical texture discrimination, we examined texture discrimination asymmetries as a function of eccentricity. We found three patterns of results; (i) both texture arrangements (A in B, and B in A) elicit a CPD but do not show an asymmetry, (ii) both texture arrangements elicit a monotonic decrease in performance with eccentricity (i.e. no CPD) but an asymmetry is seen at each eccentricity and (iii) discrimination asymmetries are minimal at fixation and in the far periphery and maximal about 3 degrees from fixation with a CPD generally shown for the 'stronger' member of the pair. These results emphasize that one cannot talk about the 'discriminability' of a particular texture pair without reference to the arrangement of the two textures and the eccentricity of presentation.     

Vibrotactile adaptation impairs discrimination of fine,but not coarse,textures

The effect of vibrotactile adaptation on the ability to discriminate textured surfaces was examined in three experiments. The surfaces were rectilinear arrays of pyramids produced by etching of silicon wafers. Adaptation to 100-Hz vibration severely hampered discrimination of surfaces with spatial periods below 100 &#119 m (Experiment 1), but had little effect on the discrimination of coarser textures (Experiment 2). To determine which vibrotactile channel—Rapidly Adapting or Pacinian—plays the larger role in mediating the discrimination of fine textures, widely separated adapting frequencies (10 and 250 Hz) were used in Experiment 3. The fact that high- but not low-frequency adaptation interfered with discrimination suggests that the Pacinian system contributes importantly to this ability. Taken as a whole, the results of this study strongly support the duplex theory of tactile texture perception, according to which different mechanisms—spatial and vibrotactile—mediate the perception of coarse and fine textures, respectively.     

On perceptual analyzers underlying visual texture discrimination: Part I

We have found a class of feature detectors, based on the quasi-collinearity of dots, which result in visual texture discrimination even when second order statistics are equal. This degenerate counterexample to the Julesz conjecture on effortless texture discrimination has supplied the key to a simple theory of texture discrimination. Accordingly, effortless texture discrimination is based on two classses of perceptual detectors: Class A, those that measure differences in second-order (dipole) statistics; Class B, those that can still detect statistical differences in some features when second-order statistics are kept identical; for instance, the quasi-collinearity of adjacent dipoles. The difference thresholds (tuning curves) for the perceptual dipole and quasi-collinearity detectors have been determined. These texture pairs were generated by a method that creates micropatterns with iso-dipole duals from 4 disks. The extension of this 4-disk method to 5 and more disks with iso-dipole duals permits the search for other kinds of perceptual detectors and will be discussed in Part II.     

Two models of the recognition and detection of texture-defined letters compared

We quantified texture segregation by measuring psychophysically the percentage correct detection scores for each of a set of 10 texture-defined (TD) letters using the temporal two-alternative forced choice method, and at the same time quantified spatial discrimination of the TD form of measuring psychophysically the percentage correct letter recognition scores for the 10 letters. Ten levels of task difficulty were created by adding noise dots to the texture patterns. The resulting psychophysical data were used to test and compare models of the detection and recognition of texture-defined letters. Each model comprised a sequence of physiologically plausible stages in early visual processing. Each had the same first, second and third stages, namely linear orientation-tuned spatial filters followed by rectification and smoothing. Model 1 had only one non-linear stage. Model 2 had two non-linear stages. In model 2 the second non-linear stage was cross-orientation inhibition. This second non-linear stage enhanced the texture borders by, in effect, comparing textures at different locations in the texture pattern. In both models, the last stage modelled either letter detection or letter recognition. Letter recognition was modelled as follows. We passed a given letter stimulus through the first several stages of a model and, in 10 separate calculations, cross-correlated the output with a template of each of the 10 letters. From these 10 correlations we obtained a predicted percentage correct letter recognition score for the given letter stimulus. The predicted recognition scores closely agreed with the experimental data at all 10 levels of task difficulty for model 2, but not for model 1. We conclude that a borderenhancing algorithm is necessary to model letter recognition. The letter-detection algorithm modelled detection of part of a letter (a single letter stroke) in terms of the signal-to-noise ratio of a letter-segment detector. The predicted letter detection scores fitted the data closely for both models.     

A texture-processing model of the ‘visual sense of number’

It has been suggested that numerosity is an elementary quality of perception, similar to colour. If so (and despite considerable investigation), its mechanism remains unknown. Here, we show that observers require on average a massive difference of approximately 40% to detect a change in the number of objects that vary irrelevantly in blur, contrast and spatial separation, and that some naive observers require even more than this. We suggest that relative numerosity is a type of texture discrimination and that a simple model computing the contrast energy at fine spatial scales in the image can perform at least as well as human observers. Like some human observers, this mechanism finds it harder to discriminate relative numerosity in two patterns with different degrees of blur, but it still outpaces the human. We propose energy discrimination as a benchmark model against which more complex models and new data can be tested.     

Asymmetries in visual texture discrimination

Recent results have shown that texture discrimination is an asymmetrical process; texture A within texture B may be much easier to detect than texture B within texture A. Two questions regarding discrimination asymmetries are addressed: (i) what sorts of textural properties are associated with discrimination asymmetries; and (ii) what sort of architecture would yield asymmetries. Two experiments show that discrimination asymmetries obtain when textures comprise circles of different sizes (large circles are easier to detect in small than vice versa) and when circles differ only in the regularity of their placement (irregularly placed circles are easier to detect in a background of regularly placed circles than vice versa). A plausible account of texture discrimination would involve the decomposition of images via a set orientation and scale selective filters followed by a second layer of filtering to detect energy differences between adjacent regions in the original convolutions. Discrimination asymmetries provide prima facie evidence against such a model because it involves only local measurements and comparisons. We propose that discrimination asymmetries are elegantly explained if it is assumed that the responses of the orientation and scale selective filters are normalized by the degree to which similarly tuned operators are responding elsewhere in the image; viz., global normalization of filter responses. However, there are cases where such global normalization is not required to explain asymmetrical discrimination.     

Figure-ground discrimination by relative movement in the visual system of the fly

The visual system of the fly performs various computations on photoreceptor outputs. The detection and measurement of movement is based on simple nonlinear multiplication-like interactions between adjacent pairs and groups of photoreceptors. The position of a small contrasted object against a uniform background is measured, at least in part, by (formally) 1-input nonlinear flicker detectors. A fly can also detect and discriminate a figure that moves relative to a ground texture. This computation of relative movement relies on a more complex algorithm, one which detects discontinuities in the movement field. The experiments described in this paper indicate that the outputs of neighbouring movement detectors interact in a multiplication-like fashion and then in turn inhibit locally the flicker detectors. The following main characteristic properties (partly a direct consequence of the algorithm's structure) have been established experimentally: a) Coherent motion of figure and ground inhibit the position detectors whereas incoherent motion fails to produce inhibition near the edges of the moving figure (provided the textures of figure and ground are similar). b) The movement detectors underlying this particular computation are direction-insensitive at input frequencies (at the photoreceptor level) above 2.3 Hz. They become increasingly direction-sensitive for lower input frequencies. c) At higher input frequencies the fly cannot discriminate an object against a texture oscillating at the same frequency and amplitude at 0° and 180° phase, whereas 90° or 270° phase shift between figure and ground oscillations yields maximum discrimination. d) Under conditions of coherent movement, strong spatial incoherence is detected by the same mechanism. The algorithm underlying the relative movement computation is further discussed as an example of a coherence measuring process, operating on the outputs of an array of movement detectors. Possible neural correlates are also mentioned.     

Gabor filters as texture discriminator

The present paper presents a model for texture discrimination based on Gabor functions. In this model the Gabor power spectrum of the micropatterns corresponding to different textures is calculated. A function that measures the difference between the spectrum of two micropatterns is introduced and its values are correlated with human performance in preattentive detection tasks. In addition, a two stage algorithm for texture segregation is presented. In the first stage the input image is transformed via Gabor filters into a representation image that allows discrimination between features by means of intensity differences. In the second stage the borders between areas of different textures are found using a Laplacian of Gaussian operator. This algorithm is sensitive to energy differences, rotation and spatial frequency and is insensitive to local translation. The model was tested by means of several simulations and was found to be in good correlation with known psychophysical characteristics as texton based texture segregation and micropattern density sensitivity. However, this simple model fails to predict human performance in discrimination tasks based on differences in the density of terminators. In this case human performance is better than expected.     

Factors limiting peripheral pattern discrimination

Previous reports indicate that some foveally discriminable compound gratings are indiscriminable in peripheral vision, even when they are scaled by the ratio of peripheral to foveal grating acuity. To determine the stimulus properties that limit peripheral discrimination, we used Gaussian derivatives of various orders. These patterns are spatially localized and have intrinsic even or odd symmetry. Our results show that certain odd symmetric patterns are discriminable in the periphery, while others are not. Furthermore, certain even symmetric patterns are not peripherally discriminable. These data are consistent with three limitations on peripheral pattern discrimination: (1) Patterns that produce different maximum neural responses will be peripherally discriminable. (2) Positional uncertainty and undersampling degrade discrimination of high spatial frequency patterns in the periphery. (3) Patterns generating substantial neural activity within a constrained region are processed as textures in peripheral vision so that pattern details within that region are no longer available for discrimination. A neural model incorporating inhibition of simple cells by complex cells implements a transition between contour analysis and texture analysis in peripheral vision and explains the experimental data.     

Learning a grating discrimination task broadens human spatial frequency tuning

 The effect of spatial frequency discrimination learning on spatial frequency detection tuning curves, obtained by a summation to threshold paradigm, has been investigated. Three human observers were exposed to a grating discrimination task for longer than two weeks, and their detection thresholds for compound Gabor gratings were measured before and after this time interval. Discrimination thresholds decreased continuously and substantially during the course of learning, while the spatial frequency detection tuning curves show significant broadening in the posttest. Calculating the discrimination resolution of an ensemble of sensory coding units shows that larger bandwidths lead to better spatial frequency discrimination performance if pattern discrimination rests on multidimensional comparison or one-dimensional scaling of the spatial frequency parameter. Further, it is shown that a multiple-mechanism nonlinear pooling model is capable of explaining the results if plasticity of coding unit bandwidth or adaptive weights of the coding unit responses at the stage of response integration is assumed. The alternative sources of plasticity and the consequences of the findings for psychophysical modeling are discussed. Received: 8 September 1999 / Accepted in revised form: 16 October 2000     

Time‐of‐Day Discriminative Learning: Contrasting the Use of Spatial Compared to Feature Information in Homing Pigeons (Columba livia)

This study examined time‐of‐day associative learning to either spatial or feature information in homing pigeons in an open‐field, laboratory setting. Homing pigeons are well known for their navigational abilities and generally have been shown to rely more heavily on spatial than nonspatial cues in recognizing a goal. However, during goal localization, homing pigeons also successfully use nonspatial, feature information. Homing pigeons were divided into two groups and were trained to locate two time‐of‐day dependent, food reward sites using either discriminative spatial or feature information. Because of the importance of the hippocampus in controlling avian memory, we hypothesized that homing pigeons trained with spatial cues would be superior in learning the time‐of‐day discrimination compared to the pigeons trained with feature cues. Indeed, homing pigeons that were trained with spatial information outperformed the pigeons trained with feature information in learning the time‐of‐day discrimination task.     

Backward masking is not required to elicit the central performance drop

In some circumstances, texture discrimination performance peaks in the parafovea rather than at the fovea. Kehrer (1987) referred to this phenomenon as the central performance drop (CPD). In most studies showing the CPD, task performance has been limited by a backward mask. Morikawa (2000) has argued that in these studies the backward mask was critical to the emergence of the CPD. In three studies we use textures comprising left and right oblique line segments and limit performance by manipulating the orientation variability within the foreground and background textures. Using this method we demonstrate that significant CPDs emerge whether or not there is a backward mask. We conclude that in past studies of the CPD the backward mask functioned primarily as a source of spatial noise and that its temporal relation to the texture display is not critical to the emergence of the CPD.     

Psychometric curve and behavioral strategies for whisker-based texture discrimination in rats

The rodent whisker system is a major model for understanding neural mechanisms for tactile sensation of surface texture (roughness). Rats discriminate surface texture using the whiskers, and several theories exist for how texture information is physically sensed by the long, moveable macrovibrissae and encoded in spiking of neurons in somatosensory cortex. However, evaluating these theories requires a psychometric curve for texture discrimination, which is lacking. Here we trained rats to discriminate rough vs. fine sandpapers and grooved vs. smooth surfaces. Rats intermixed trials at macrovibrissa contact distance (nose >2 mm from surface) with trials at shorter distance (nose <2 mm from surface). Macrovibrissae were required for distant contact trials, while microvibrissae and non-whisker tactile cues were used for short distance trials. A psychometric curve was measured for macrovibrissa-based sandpaper texture discrimination. Rats discriminated rough P150 from smoother P180, P280, and P400 sandpaper (100, 82, 52, and 35 μm mean grit size, respectively). Use of olfactory, visual, and auditory cues was ruled out. This is the highest reported resolution for rodent texture discrimination, and constrains models of neural coding of texture information.     

Some task and signal dependent rules for spatial vision

In this paper we consider the types of computational processes which may be involved in solving a variety of perceptual problems from the detection of signals in the presence of others, to texture discrimination, and some aspects of pattern recognition. These processes centre around the involvement of correlational computations, the transduction of their input/output values, and the apparent involvement of selective filtering mechanisms. Our results suggest that even if fixed detectors (in tuning characteristics) are involved in low-level vision, the human observer apparently employs much more adaptive (variable tuning characteristic) filters and nonlinear mechanisms in more complex spatial tasks.     

Models for preattentive texture discrimination: Fourier analysis and local feature processing in a unified framework

Spatial frequency analysis and local feature analysis may be considered to be examples of a class of models for texture discrimination. In this theoretical framework, texture discrimination relies on differences in the distribution of responses generated in linear receptive fields placed randomly on the texture. If the set of receptive fields is taken to be a collection of gratings, spatial-frequency analysis is recovered. If the set of receptive-fields is taken to be a collection of local feature templates, a corresponding local-feature model is recovered. In order to test such models, it is necessary to construct distinct texture pairs that elicit similar distributions of responses for all of the postulated receptive field profiles: the model prediction is that such textures are not discriminable. A method is provided for construction of such textures which test generic models within this framework. This framework includes not only strict Fourier analysis, but also models which postulate a collection of arbitrarily-shaped local feature detectors, and models which postulate both Fourier analysis and local feature detection.     

Orientation variance as a quantifier of structure in texture

I consider how structure is derived from texture containing changes in orientation over space, and propose that multi-local orientation variance (the average orientation variance across a series of discrete images locales) is an estimate of the degree of organization that is useful both for spatial scale selection and for discriminating structure from noise. The oriented textures used in this paper are Glass patterns, which contain structure at a narrow range of scales. The effect of adding noise to Glass patterns, on a structure versus noise task (Maloney et al., 1987), is compared to discrimination based on orientation variance and template matching (i.e. having prior knowledge of the target's orientation structure). At all but very low densities, the variance model accounts well for human data. Next, both models' estimates of tolerable orientation variance are shown to be broadly consistent with human discrimination of texture from noise. However, neither model can account for subjects' lower tolerance to noise for translational patterns than other (e.g. rotational) patterns. Finally, to investigate how well these structural measures preserve local orientation discontinuities, I show that the presence of a patch of unstructured dots embedded in a Glass pattern produces a change in multi-local orientation variance that is sufficient to account for human detection (Hel Or and Zucker, 1989). Together, these data suggest that simple orientation statistics could drive a range of 'texture tasks', although the dependency of noise resistance on the pattern type (rotation, translation, etc.) remains to be accounted for.     

Visual textures, machine vision and animal camouflage

The psychophysics of visual texture perception and texture discrimination have been investigated extensively during the past 30 years. Humans have been the main study subjects, but some research on texture perception has involved other species, and there is good reason to think that the most general results from humans apply to other vertebrates as well. Psychophysicists have suggested that some of their findings on human vision reflect adaptive 'tricks' for countering prey camouflage, but this possibility has not been widely communicated to evolutionary biologists. We review the psychophysicists' main conclusions on texture discrimination, and list additional questions that their results raise when animal coats are considered as visual textures. We also suggest ways in which advances in computer vision can be combined with psychophysics to provide new perspectives on the function of animal coat patterns.