Bela Julesz

Visual Pattern Discrimination

Visual discrimination experiments were conducted using unfamiliar displays generated by a digital computer. The displays contained two side-by-side fields with different statistical, topological or heuristic properties. Discrimination was defined as that spontaneous visual process which gives the immediate impression of two distinct fields. The condition for such discrimination was found to be based primarily on clusters or lines formed by proximate points of uniform brightness. A similar rule of connectivity with hue replacing brightness was obtained by using varicolored dots of equal subjective brightness. The limitations in discriminating complex line structures were also investigated.

Parallel versus serial processing in rapid pattern discrimination

When stimuli are available for just a brief period (∼100 ms) only restricted spatial information can be processed by the visual system. If the stimuli are presented very briefly, eye movements are not possible. The time during which the after-image of the stimulus is available for inspection is terminated by presentation of a masking pattern. We show here that in these conditions a small pattern is easily detected against a background made up of many others, only if this target pattern differs from the background patterns in certain local features. In this case the detectability of the target is almost independent of the number of background elements, suggesting that a parallel process is operating. Detection of patterns not differing from their backgrounds in such features requires focal attention which is a serial process. The aperture of this attention is scaled to minimize the number of shifts of attention required.

Binocular Depth Perception without Familiarity Cues Random-dot stereo images with controlled spatial and temporal properties clarify problems in stereopsis

The reported phenomena were obtained through the use of special techniques. (i) All monocular depth and familiarity cues were removed from the stimuli (through the use of randomdot stereo patterns). (ii) The statistical and topological properties of the stimuli were precisely known (since they were generated according to a specific computer program). (iii) Convergence motions of the eye and proprioceptive cues were eliminated (through the use of tachistoscopic illumination). (iv) The time of presentation was under control (through erasure of the persistent afterimages). Under these conditions stereopsis could be studied in its purest form. It was shown that depth can be perceived in the absence of monocular depth and familiarity cues and of all binocular depth cues except for disparity. These findings have important implications for some existing theories of stereopsis and open up areas for further research. Some phenomena based on stereo erasure are reported here for the first time. It has been demonstrated that the perception of ambiguous depth organizations can be influenced, even subliminally, by a preceding unambiguous stimulus. Perhaps the most interesting result is the finding that the correspondence of objects and patterns in the two retinal projections can be established without actual recognition of the objects and patterns. This pattern matching is based on some relatively simple processes of finding connected clusters formed by adjacent points of similar brightness, and the processes seem to be amenable to rigorous analysis.

Spatial-Frequency Masking in Vision: Critical Bands and Spread of Masking*

Vertical sinusoidal gratings were viewed in masking noise consisting of vertical stripes spread along the horizontal direction. Masking functions were obtained while varying the grating frequency relative to various one-octave-wide bands of noise. These functions closely resemble curves derived from previous experiments on adaptation to gratings. Masking was also measured as a function of the width of a band of noise centered on the grating frequency. Masking increased as the band was widened up to approximately ±1 octave; masking did not increase further when the band was widened beyond this range. The results demonstrate that a grating is masked only by noise whose spatial frequencies are similar to the grating frequency. The experiments provide further indication of the existence of channels in the visual system that are selectively tuned to different spatial frequencies.

Texton gradients : The texton theory revisited

A brief outline of the texton theory was given in several review papers (Julesz and Bergen 1983; Julesz 1984a, b, 1985) without going into details. Here a more complete version of the texton theory is presented, with emphasis on the critical distances within which the density of textons is determined by the preattentive system. Particularly some recent findings by Sagi and Julesz (1985a, b) influenced the current version of the texton theory. The stimuli are restricted to drawing composed of line segments that permit a precise definition of neighborhoods and distances to which the line segments and texture elements have to be confined in order to quality for preattentive texture discrimination. These critical distances and the aperture of focal attention are scaled by the average size of the texture elements. Furthermore, it is stressed that even when the stimuli are restricted to line segments, the blobs outlined by line segments behave like textons. The preattentive system ignores the exact shape of these blobs, but is sensitive to their average width, length, and orientation.

Inability of Humans to Discriminate between Visual Textures That Agree in Second-Order Statistics—Revisited

In an earlier study by Julesz (1962) pairs of random textures were generated side-by-side using a Markov process with different third-order joint-probability distributions but identical first- and second-order distributions. Such texture pairs could not be discriminated from each other by the human visual system without scrutiny. Unfortunately, Markov processes are inherently one-dimensional while the general processes underlying visual texture discrimination are two-dimensional. Here three new methods are introduced that generate two-dimensional non-Markovian textures with different third-order but identical first- and second-order statistics. All three methods generate texture pairs that cannot be discriminated from each other. The lack of texture discrimination is the more astonishing since the individual elements that form the texture pair are clearly perceived as being very different. However, a counterexample was found that yields discrimination although the texture pair has approximately identical second-order statistics. This case can be explained by assuming that early feature extractors do some preprocessing. These new demonstrations give support to a model of texture discrimination in which the stimulus is first analyzed by local feature extractors that can detect only simple features such as dots and edges of given sizes and orientations. Then the outputs of these simple extractors are evaluated by a global processor that can compute only second- or first-order statistics (that is can compare at most two such outputs).

Masking in Visual Recognition: Effects of Two-Dimensional Filtered Noise

It is difficult to recognize portraits that have been coarsely sampled and quantized. Blurring such images improves recognition. A simple, straightforward explanation is that high-frequency noise introduced by the sampling and quantizing must be removed by low-pass filtering to improve the signal-to-noise ratio and hence signal detectability or recognition. Experiments reported here, suggested on the basis of a different model, show instead that noise bands that are spectrally adjacent to the pictures spectrum are considerably more effective in suppressing recognition.

A brief outline of the texton theory of human vision

Visual perception operates in a preattentive mode and in an attentive mode. In the preattentive mode no complex forms are processed, and yet in parallel, without effort or scrutiny, differences in a few local conspicuous features (called textons) are detected over the entire visual field. It is these loci of texton differences to which the narrow aperture of attention is directed in steps each lasting about 50 ms. Only in this aperture of attention are the positional relations between textons preserved, permitting form recognition.

Independent Spatial-Frequency-Tuned Channels in Binocular Fusion and Rivalry

Monocular masking studies show that the visibility of a one-dimensional sinusoidal grating remains unchanged in the presence of masking noise filtered so as to contain spectral components that are at least two octaves away from the spatial frequency of the grating (Stromeyer and Julesz 1972). In the present study, random-dot stereograms were bandpass filtered in the two-dimensional Fourier domain, and masking noise of various spatial frequency bands was added to the filtered stereograms. Masking noise bands containing equally effective noise energy were selected such that their bands were either overlapping with the stereoscopic image spectrum or were two octaves distant. The first case resulted in binocular rivalry; however, in the second case stereoscopic fusion could be maintained in the presence of strong binocular rivalry owing to the masking noise. This finding indicates that spatial-frequency-tuned channels are not restricted to one-dimensional gratings but operate on two-dimensional patterns as well. Furthermore, these frequency channels are utilized in stereopsis and work independently from each other, since some of these channels can be in binocular rivalry while at the same time other channels yield fusion. The main binocular experiments are demonstrated.

Visual discrimination of textures with identical third-order statistics

We found a new class of two-dimensional random textures with identical third-order statistics that can be effortlessly discriminated. Discrimination is based on local “granularity” differences between these iso-trigon texture pairs. This is the more surprising since it is commonly assumed that texture granularity (grain) is determined by the power spectrum which, in turn, can be obtained from the second-order statistics. Because textures with identical third-order statistics must have identical second-order statistics (i.e., identical power spectra), visible texture granularity is not controlled by power spectra, and not even by third-order statistics.