Perception

By the 1960s it was evident from neuroanatomy that there were extensive recurrent interactions, both excitatory and inhibitory, among visual cortical neurons. Nevertheless, the psychophysical discovery of ‘spatial-frequency channels’ gave rise to a decade in which parallel, independent channels were thought to subserve early spatial vision. Recent work, however, has clearly demonstrated that early visual channels do not perform a Fourier or wavelet decomposition of the image. Instead, they interact through a variety of nonlinear pooling mechanisms. Such nonlinear interactions perform important computations in texture perception, stereopsis, and motion and form vision.

The aim of the experiments was to discover whether the visual system has independent access to motion information at different spatial scales when presented with a broadband stimulus. Subjects were required to discriminate between a pair of two-frame motion sequences, one containing a coherently displacing pattern and the other containing a pattern with high-frequency noise. The stimuli were either narrowband (1 octave) or broadband (6 octaves spanning 0.23–15.0 cycles deg⁻¹) and their power spectra were either flat or followed a 1/f² function. For the broadband stimuli, noise was introduced cumulatively into increasingly lower frequencies. For the narrowband stimuli, noise was introduced into the same frequency band as the signal. All stimuli could be defined by the lowest noise frequency (n₁) they contained. For each stimulus, the largest spatial displacement across the two frames at which the task could be performed was measured (d_max). For the narrowband stimuli, d_max increased as n₁ was lowered. This was true over the entire frequency range for the 1/f² stimuli, though the task became impossible for the flat-spectrum stimuli at the lowest frequencies. This is attributed to the very low contrast of these latter stimuli. The d_max values for the broadband stimuli tended to shadow those of the narrowband stimuli with the equivalent values of n₁ being around 25% lower. The data were modelled by spatiotemporally filtering the stimuli and considering the amount of directional power in the signal and noise sequences. The results suggest that there must be multiple spatial-frequency channels in operation, and that for broadband patterns the visual system has perceptual access to these individual channel outputs, utilising different filters depending on the task requirements.

How are binocular disparities encoded and represented in the human visual system? An ‘encoding cube’ diagram is introduced to visualise differences between competing models. To distinguish the models experimentally, the depth-increment-detection function (discriminating disparity d from d ± Δd) was measured as a function of standing disparity (d) with spatially filtered random-dot stereograms of different centre spatial frequencies. Stereothresholds degraded more quickly as standing disparity was increased with stimuli defined by high rather than low centre spatial frequency. This is consistent with a close correlation between the spatial scale of detection mechanisms and the disparities they process. It is shown that a simple model, where discrimination is limited by the noisy ratio of outputs of three disparity-selective mechanisms at each spatial scale, can account for the data. It is not necessary to invoke a population code for disparity to model the depth-increment-detection function.

This type of encoding scheme implies insensitivity to large interocular phase differences. Might the system have developed a strategy to disambiguate or shift the matches made at fine scales with those made at the coarse scales at large standing disparities? In agreement with Rohaly and Wilson, no evidence was found that this is so. Such a scheme would predict that stereothresholds determined with targets composed of compounds of high and low frequency should be superior to those of either component alone. Although a small stereoacuity benefit was found at small disparities, the more striking result was that stereothresholds for compound-frequency targets were actually degraded at large standing disparities. The results argue against neural shifting of the matching range of fine scales by coarse-scale matches posited by certain stereo models.

Human subjects can perceive global motion or motions in displays containing diverse local motions, implying representation of velocity at multiple scales. The phenomena of flexible global direction judgments, and especially of motion transparency, also raise the issue of whether the representation of velocity at any one scale is single-valued or multi-valued. A new performance-based measure of transparency confirms that the visual system represents directional information for each component of a transparent display. However, results with the locally paired random-dot display introduced by Qian et al, show that representations of multiple velocities do not coexist at the finest spatial scale of motion analysis. Functionally distinct scales of motion processing may be associated with (i) local motion detectors which show a strong winner-take-all interaction; (ii) spatial integration of local signals to disambiguate velocity; (iii) selection of reliable velocity signals as proposed in the model of Nowlan and Sejnowski; (iv) object-based or surface-based representations that are not necessarily organised in a fixed spatial matrix. These possibilities are discussed in relation to the neurobiological organisation of the visual motion pathway.

Thresholds were measured for discriminating changes in the slopes of the amplitude spectra of stimuli derived from photographs of natural scenes and from random-luminance patterns. The variety and magnitudes of the thresholds could be explained by a model based on the discrimination of the changes in band-limited local contrast. Different spatial scales of local contrast (or different spatial-frequency bands of about 1 octave) were implicated for different reference spectral slopes; the model implicated a lower frequency-band for stimuli with shallower amplitude spectra. The implications of the model were tested experimentally by using stimuli in which the spectra were changed within restricted spatial-frequency bands. When the amplitude spectra of the test and reference stimuli differed only within the implicated frequency bands, thresholds were affected little. However, when the test and reference spectra differed at all frequencies except those in the implicated bands, the thresholds were elevated markedly. The forms of the psychometric functions for the discrimination task were entirely compatible with the hypothesis that the task relies upon the ability to discriminate changes of contrast. The Weibull functions fitted to the data had slope parameters (β) in the range 1 to 3, compatible with discrimination of low (but suprathreshold) contrasts.

Different classifications of an identical visual stimulus may require different perceptual properties from the visual input. How do processes of object and scene categorisation use the information associated with different perceptual spatial scales? One scenario suggests that recognition should use coarse blobs before fine-scale edges because scale usage is perceptually determined. However, perceptual determination neglects one important aspect of any recognition task: the information demands of the considered classification of the input. Evidence is reviewed suggesting that scale usage could be flexibly determined by the diagnosticity of scale-specific cues for different categorisations of scenes and faces.

Studies of the microgenesis of perception led to the hypothesis that global aspects of objects are processed faster than their details. If one starts with the assumption that low-frequency information of objects corresponds to the global, and high-frequency information to the local aspects, recognising objects should rely at first on information from the low-frequency channels and afterwards from that of the high-frequency channels. The priming paradigm provides a mean of investigating experimentally the temporal availability of low-frequency and high-frequency information in object perception. In the experiments subjects had to respond to target objects preceded either by related or by unrelated priming stimuli, which consisted of low-pass-filtered and high-pass-filtered versions of the objects. With the influence of stimulus complexity controlled, pictures of objects were chosen that varied in the number of intensity changes in the high-frequency components, with those of the low-frequency components kept constant. The exposure duration of each prime was varied between 40 and 100 ms. The results indicated that target identification only profits more from low-frequency than from high-frequency primes if the high-frequency information has a high level of complexity. If the number of intensity changes in the high-frequency components of the prime is low, target identification is most strongly facilitated. The results are discussed in terms of models which focus on organising principles at different scales.

It has previously been observed that low spatial frequencies (≤ 1.0 cycles deg⁻¹) tend to dominate high spatial frequencies (≥ 5.0 cycles deg⁻¹) in several types of visual-information-processing tasks. This earlier work employed reaction times as the primary performance measure and the present experiments address the possibility of low-frequency dominance by evaluating visually guided performance of a completely different response system: the control of slow-pursuit eye movements. Slow-pursuit gains (eye velocity/stimulus velocity) were obtained while observers attempted to track the motion of a sine-wave grating. The drifting gratings were presented on three types of background: a uniform background, a background consisting of a stationary grating, or a flickering background. Low-frequency dominance was evident over a wide range of velocities, in that a stationary high-frequency component produced little disruption in the pursuit of a drifting low spatial frequency, but a stationary low frequency interfered substantially with the tracking of a moving high spatial frequency. Pursuit was unaffected by temporal modulation of the background, suggesting that these effects are due to the spatial characteristics of the stationary grating. Similar asymmetries were observed with respect to the stability of fixation: active fixation was less stable in the presence of a drifting low frequency than in the presence of a drifting high frequency.

Measurements were carried out of saccadic eye movements made during brief (3 s) examination of images which the observer was asked to identify. Each image was identified in three forms: low-pass filtered, high-pass filtered, and unfiltered. The analysis of the eye-movement patterns was based on the locations of fixations made during examination of the images, for which purpose a least-squares measure of similarity between two sets of locations was introduced. It is shown that there is a high degree of similarity between fixations made by the same observer to the different versions of a given image and that for a given image there is a high degree of similarity between fixations made by the eighteen observers who participated in the experiments. The similarities are greater for the initial 1.5 s than for the full viewing period of 3 s. The similarity between the locations of fixations and those of selected image features such as local contrast, high-spatial-frequency content, and edge density was also examined. It is shown that there is only weak similarity between the locations of fixations and those of any given local image feature, and the tendency of observers to fixate centrally on the image is identified as the principal reason for the low similarity values. It is shown that if the nonuniform distribution of eye movements is taken into account, significant similarities are found between the locations of fixations and those of certain image features, such as edge density.