Dov Sagi

Lateral interactions between spatial channels: Suppression and facilitation revealed by lateral masking experiments

We measured contrast detection thresholds for a foveal Gabor signal flanked by two high contrast Gabor signals. The spatially localized target and masks enabled investigation of space dependent lateral interactions between foveal and neighboring spatial channels. Our data show a suppressive region extending to a radius of two wavelengths, in which the presence of the masking signals have the effect of increasing target threshold. Beyond this range a much larger facilitatory region (up to a distance of ten wavelengths) is indicated, in which contrast thresholds were found to decrease by up to a factor of two. The interactions between the foveal target and the flanking Gabor signals are spatial-frequency and orientation specific in both regions, but less specific in the suppression region.

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.

The architecture of perceptual spatial interactions

Lateral interactions between spatial filters were explored with a lateral masking paradigm. Contrast sensitivity (two-alternative forced-choice) for a Gabor signal in the presence of two flanking high contrast Gabor signals (masks) was measured. When the target to mask distance was less than 2 target wavelengths the contrast sensitivity decreased up to a factor of two relative to a no mask condition. At larger separations, up to eight wavelengths, an increase in contrast sensitivity occurred. This increase was maximal at separation distances of 2-3 wavelengths, where sensitivity increased by a factor of two. However, the enhancement magnitude and range was dependent on the offset between the Gabor signal orientation and the direction defined by the virtual line connecting the two masks (global orientation). Maximal effects occurred when this offset was zero (100% increase in sensitivity) and 90 deg (50% increase). A 45 deg offset yielded only a small enhancement (20%). The enhancement dependence on spatial arrangement was found to be invariant across different global orientations (meridian). This pattern of interactions may be involved in grouping colinear line segments into smooth curves.

Motion-induced blindness in normal observers

Cases in which salient visual stimuli do not register consciously are known to occur in special conditions, such as the presentation of dissimilar stimuli to the two eyes or when images are stabilized on the retina. Here, we report a striking phenomenon of ‘visual disappearance’ observed with normal-sighted observers under natural conditions. When a global moving pattern is superimposed on high-contrast stationary or slowly moving stimuli, the latter disappear and reappear alternately for periods of several seconds. We show that this motion-induced blindness (MIB) phenomenon is unlikely to reflect retinal suppression, sensory masking or adaptation. The phenomenology observed includes perceptual grouping effects, object rivalry and visual field anisotropy. This is very similar to that found in other types of visual disappearance, as well as in clinical cases of attention deficits, in which partial invisibility might occur despite the primary visual areas being intact. Disappearance might reflect a disruption of attentional processing, which shifts the system into a winner-takes-all mode, uncovering the dynamics of competition between object representations within the human visual system.

Vision outside the focus of attention

We investigated the relationship between focal attention and a feature-gradient detection that is performed in a parallel manner. We found that a feature gradient can be detected without measurable impairment of performance even while a concurrent form-recognition task is carried out, in spite of the fact that the form-recognition task engages focal attention and thus removes attentive resources from the vicinity of the feature gradient. This outcome suggests strongly that certain perceptions concerning salient boundaries and singularities in a visual scene can be accomplished without the aid of resource-limited processes, such as focal attention, and, by implication, that there may exist two distinct perceptual faculties (one attentive, the other not) that are able to bring complementary kinds of visual information simultaneously to our awareness.

Perceptual learning: learning to see

Perceptual learning in vision has been found to be highly specific for simple stimulus attributes, implying highly specific modifications in the nervous system. The type of specificity found (location, orientation, eye) implied plasticity at very early stages of visual processing, where processing modules were believed to be hard-wired and task independent. Recent studies show, however, that learning is task dependent. Studies examining the time course of learning indicate that at least two different learning processes are involved in perceptual learning, reflecting different levels of processing. Perceptual learning appears to be governed by associative rules and to be constrained by system architecture.

Short-range limitation on detection of feature differences.

We studied the ability of observers to detect the presence of a clearly visible line segment against a background of line segments of different orientation. As we increase the number (density) of these background lines, we find that detectability does not behave monotonically. Adding a small number of background lines decreases detectability but if adjacent line segments are permitted to fall in close range, a further increase of background lines improves performance which eventually reaches a constant level. This suggests that detection of feature differences involves a short-range process. The range of this process is about two degrees or twice the length of the line segments used. Thus texture-gradients between different elements are only formed if the distance between these elements is not much larger than the average element size.

Detection versus discrimination of visual orientation

The role of focused attention in vision is examined. Recent theories of attention hypothesize that serial search by focal attention is required for discrimination between different combinations of features. Experiments are reported which show that the mixture of a few (less than five) horizontal and vertical line segments embedded in an aggregate of diagonal line segments can be rapidly counted (also called ‘subitizing’) by a parallel (preattentive) process, while the discrimination between horizontal and vertical orientation requires serial search by shifting focal attention to each line segment. Thus detecting and counting targets that differ in orientation can be done in parallel by a preattentive process, whereas knowing ‘what’ the orientation of a target is (horizontal or vertical, ie of a single conspicuous feature) requires a serial search by focal attention.

Isolating excitatory and inhibitory nonlinear spatial interactions involved in contrast detection

Interactions between filters tuned to different orientations and spatial locations were investigated with a masking paradigm. Targets were masked by pairs of Gabor signals presented either at a different orientation (+/- delta theta) or at a different spatial location (+/- delta y). The two mask components were either of equal phase or of opposite phase to each other. Detection thresholds of the target were measured as a function of mask contrast. Typically, the curves obtained showed the following behavior: for increasing mask contrast the threshold first decreased, then reached a minimum and then increased linearly on a log-log scale reflecting a power-law behavior. Mask pairs of equal phase as well as pairs of opposite phase were shown to facilitate detection. Facilitation by mask pairs of equal phase was larger (up to 0.4 log units) and decreased for increasing delta theta and delta y. The facilitation for mask pairs of opposite phase (approximately 0.1 log units) was observed only for larger delta theta and delta y. Phase independent suppression was observed with higher mask contrasts at smaller delta theta and delta y. The strength of this suppression was shown to decrease with practice. We account for the observed facilitation with an accelerating transducer function applied on a second-stage filter. Suppression is modeled with an additional inhibitory second stage filter that divides the output of this transducer. Selective reduction of the inhibitory gain accounts for the practice effects.

Spatial variability as a limiting factor in texture-discrimination tasks: implications for performance asymmetries

Texture-discrimination tasks reveal a pronounced performance asymmetry depending on which texture represents the foreground region (small area) and which represents the ground (large area). This asymmetry implies that some global processes are involved in the segmentation process. We examined this problem within the context of the texture-segmentation algorithm, assuming two filtering stages. The first stage uses spatial frequency and orientation-selective (Gabor) filters, whereas the second stage is formed by low-resolution edge-detection filters. The presence and location of texture borders are indicated by significant responses in the second stage. Spurious texture borders may occur owing to textural local variabilities (such as orientation randomization), which are enhanced by the first stage. We suggest that these spurious borders act as background noise and thus limit performance in texture-discrimination tasks. The noise level depends on which texture occupies the ground in the display. We tested this model on numerous pairs of textures and found remarkably good correlation with human performance. A prediction of the model, namely, that discrimination asymmetry will be reduced when textural elements have identical orientation, was tested psychophysically and confirmed.