Segmentation versus integration in visual motion processing

doi:10.1016/0166-2236(93)90179-P

Trends in Neurosciences

Volume 16, Issue 7, July 1993, Pages 263-268

https://doi.org/10.1016/0166-2236(93)90179-P Get rights and content

Abstract

Reliable motion perception requires processes that integrate visual motion signals from neighbouring locations in the visual field, which should have the effect of smoothing out spatial variations in velocity. However, we also require motion processing to be very sensitive to local velocity differences, so that moving objects appear sharply distinct from their background and specific differential properties of optic flow associated with the observer's motion can be detected. Perceptual experiments give evidence both for integrative processes, which lead to spreading of perceived motion, and for differential processes, which lead to motion contrast and segmentation. Current and future experiments might allow tests of theoretical schemes that employ adaptive networks and/or multiple representations in order to reconcile the conflicting demands of integration and segmentation.

References (58)

B.K.P. Horn et al.
Artif. Intell.
(1981)
O.J. Braddick
Vision Res.
(1974)
D. Regan et al.
Vision Res.
(1978)
K.I. Beverley et al.
Vision Res.
(1979)
P. Tynan et al.
Vision Res.
(1975)
M. Nawrot et al.
Vision Res.
(1990)
E.C. Hildreth
The Measurement of Visual Motion
(1984)
A. Yuille et al.
Nature
(1988)
E.H. Adelson et al.
Nature
(1982)
B. Julesz
Foundations of Cyclopean Perception
(1971)

S. Ullman

The Interpretation of Visual Motion

(1979)

W. Reichardt et al.

Naturwissenschaften

(1988)

S. Uras et al.

Biol. Cybern.

(1988)

M.V. Srinivasan

Biol. Cybern.

(1990)

T. Poggio et al.

S.M. Anstis

Vision Res.

(1970)

B.J. Frost et al.

Science

(1983)

J. Allman et al.

Perception

(1985)

J. Allman et al.

Annu. Rev. Neurosci.

(1985)

W. Reichardt et al.

Biol. Cybern.

(1979)

T. Poggio et al.

Naturwissenschaften

(1981)

M. Egelhaaf

Biol. Cybern.

(1985)

H.C. Longuet-Higgins et al.

J.J. Koenderink

Vision Res.

(1986)

A. Blake et al.

Visual Reconstruction

(1987)

J.J. Koenderink et al.

J. Opt. Soc. Am.

(1976)

K. Nakayama

B.C.J. Rogers

Nature

(1988)

H. Saito

J. Neurosci.

(1986)

Cited by (205)

Adaptive mechanisms of visual motion discrimination, integration, and segregation
2021, Vision Research
Under ecological conditions, the luminance impinging on the retina varies within a dynamic range of 220 dB. Stimulus contrast can also vary drastically within a scene and eye movements leave little time for sampling luminance. Given these fundamental problems, the human brain allocates a significant amount of resources and deploys both structural and functional solutions that work in tandem to compress this range. Here we propose a new dynamic neural model built upon well-established canonical neural mechanisms. The model consists of two feed-forward stages. The first stage encodes the stimulus spatially and normalizes its activity by extracting contrast and discounting the background luminance. These normalized activities allow a second stage to implement a contrast-dependent spatial-integration strategy. We show how the properties of this model can account for adaptive properties of motion discrimination, integration, and segregation.
Redundancy between spectral and higher-order texture statistics for natural image segmentation
2021, Vision Research
Visual texture, defined by local image statistics, provides important information to the human visual system for perceptual segmentation. Second-order or spectral statistics (equivalent to the Fourier power spectrum) are a well-studied segmentation cue. However, the role of higher-order statistics (HOS) in segmentation remains unclear, particularly for natural images. Recent experiments indicate that, in peripheral vision, the HOS of the widely adopted Portilla-Simoncelli texture model are a weak segmentation cue compared to spectral statistics, despite the fact that both are necessary to explain other perceptual phenomena and to support high-quality texture synthesis. Here we test whether this discrepancy reflects a property of natural image statistics. First, we observe that differences in spectral statistics across segments of natural images are redundant with differences in HOS. Second, using linear and nonlinear classifiers, we show that each set of statistics individually affords high performance in natural scenes and texture segmentation tasks, but combining spectral statistics and HOS produces relatively small improvements. Third, we find that HOS improve segmentation for a subset of images, although these images are difficult to identify. We also find that different subsets of HOS improve segmentation to a different extent, in agreement with previous physiological and perceptual work. These results show that the HOS add modestly to spectral statistics for natural image segmentation. We speculate that tuning to natural image statistics under resource constraints could explain the weak contribution of HOS to perceptual segmentation in human peripheral vision.
A neural index of inefficient evidence accumulation in dyslexia underlying slow perceptual decision making
2021, Cortex
Citation Excerpt :
Noise exclusion is therefore necessary to form a perceptual filter and isolate the signal dots to detect motion (Dakin et al., 2005). Both integration and segregation have been recognised as competing visual forces previously (Braddick, 1993; Watamaniuk et al., 2003) and have been proposed to have distinct neural mechanisms (McDonald et al., 2013). Thus, it may not be the integration of motion information that is dysfunctional in dyslexia but rather an inability to segregate noise from signal information (Conlon et al., 2013).
Visual processing deficits have been widely reported in developmental dyslexia however the locus of cognitive dysfunction remains unclear. Here, we examined the neural correlates of perceptual decision-making using a dot-motion task and electroencephalography (EEG) and investigated whether presenting deficits were unique to children with dyslexia or if they were also evident in other, typically developing children with equally immature reading systems. Sixty-eight children participated: 32 with dyslexia (DD; 16 females); 21 age-matched controls (AM; 11 females) and 15 reading-matched controls (RM; 9 females). All participants completed a bilaterally presented random-dot-motion task while EEG was recorded. Neural signatures of low level sensory processing (steady state visual evoked potentials; SSVEPs), pre-target attentional bias (posterior α power), attentional orienting (N2), evidence accumulation (centro-parietal positive decision signal; CPP) and execution of a motor response (β) were obtained to dissect the temporal sequence of perceptual decision-making. Reading profile provided a score of relative lexical and sublexical skills for each participant. Although all groups performed comparably in terms of task accuracy and false alarm rate, the DD group were slower and demonstrated an earlier peak latency, reduced slope and lower amplitude of the CPP compared with both AM and RM controls. Reading profile was found to moderate the relationship between word reading ability, reaction time as well as CPP indices showing that lexical dyslexics responded more slowly and had a shallower slope, reduced amplitude and earlier latency of CPP waveforms than sublexical dyslexics. These findings suggest that children with dyslexia, particularly those with relatively poorer lexical abilities, have a reduced rate and peak of evidence accumulation as denoted by CPP markers yet remain slow in their overt response. This is in keeping with hypotheses that children with dyslexia have impairment in effectively sampling and processing evidence about visual motion stimuli.
No effect of feedback, level of processing or stimulus presentation protocol on perceptual learning when easy and difficult trials are interleaved
2020, Vision Research
Citation Excerpt :
By contrast, receptive fields in higher cortical areas such as V3, V4 and V5, are much larger, and pool the incoming sensory information from lower level receptive fields (Burr & Thompson, 2011; Gilbert et al., 2001) across features such as space, time, speed, direction of motion and spatial frequency (Burr & Thompson, 2011; Gilbert, Sigman, & Crist, 2001; Felleman & Van Essen, 1987; Furlan & Smith, 2016; Hubel & Wiesel, 1965; Mikami, Newsome, & Wurtz, 1986; Movshon, Thompson, & Tolhurst, 1978; Sillito, Cudeiro, & Jones, 2006; Zeki, 1974). Local and global tasks have been developed in order to selectively assess these levels of processing (Newsome et al., 1989; Braddick, 1993; Gilbert et al., 2001; Bex & Dakin, 2002; Ostwald, Lam, Li, & Kourtzi, 2008; Amano, Edwards, Badcock, & Nishida, 2009; Burr & Thompson, 2011; Nishida, 2011). It should be noted that formal definitions of tasks for “local” and “global” processing do not exist, and the distinction between the two is not always clear cut, and therefore we propose these working definitions of local and global processing tasks.
The role of feedback during training is a topic of great theoretical importance in perceptual learning. Feedback can be provided externally by the environment or internally by the observer. In order to evaluate the effectiveness of learning with internal versus external feedback, we performed a large multi-level experiment, varying the type of training task (Motion or Form), the level of processing (Local or Global), the presence of feedback (With or Without) and finally the method of stimulus presentation (Adaptive staircase or Method of constant stimuli). 140 participants were randomly assigned to one of ten groups and undertook 3 days of training in one condition only. Detection thresholds were measured daily before and after training with a pre- and post-assessment. A 75% detection threshold was calculated and used to estimate that day’s training levels (65% and 85% accuracy for difficult and easy trials respectively). The group trained with MOCS were presented with predefined randomly interleaved easy and difficult trials ranging from 50% to 95% stimulus intensity. Our findings indicate that improvement was generally robust across training-tasks, processing levels and feedback conditions. This suggests that internal reinforcement is as effective as external feedback in a discrete-noise-paradigm for local and global tasks when easy and difficult trials are interleaved.
Sensory loss due to object formation
2020, Vision Research
Citation Excerpt :
For motion, it is well accepted that some form of segmentation occurs before motion information is integrated within segmented objects. Clearly, it would make no sense to integrate motion information across two distinct objects because these different objects are free to move in different directions and different speeds (Braddick, 1993; Masson & Perrinet, 2012). In comparison to the motion literature, the importance of segmentation in scene analysis is encountered less often in studies on binocular vision.
The precision to locate individual features in depth can often be improved by integrating information over space. However, this integration can sometimes be extremely detrimental, as for example in the case of the Westheimer-McKee phenomenon where features are grouped to form an object. We replicate here the known loss of precision in this phenomenon and document an additional loss of accuracy. These detrimental effects are still present when the object is elicited by other principles of organization, including a cross-modal auditory cue. Similar effects of object formation are found on lateral motion sensitivity. We then present a simple probabilistic model based on the integration of estimated depth within an object and propagation of object mean depth and uncertainty back to the elementary features of the object. This propagation of object uncertainty is a hitherto underestimated side-effect of object formation.
Visual Motion Processing in Macaque V2
2018, Cell Reports
In the primate visual system, direction-selective (DS) neurons are critical for visual motion perception. While DS neurons in the dorsal visual pathway have been well characterized, the response properties of DS neurons in other major visual areas are largely unexplored. Recent optical imaging studies in monkey visual cortex area 2 (V2) revealed clusters of DS neurons. This imaging method facilitates targeted recordings from these neurons. Using optical imaging and single-cell recording, we characterized detailed response properties of DS neurons in macaque V2. Compared with DS neurons in the dorsal areas (e.g., middle temporal area [MT]), V2 DS neurons have a smaller receptive field and a stronger antagonistic surround. They do not code speed or plaid motion but are sensitive to motion contrast. Our results suggest that V2 DS neurons play an important role in figure-ground segregation. The clusters of V2 DS neurons are likely specialized functional systems for detecting motion contrast.

View all citing articles on Scopus

View full text

Trends in Neurosciences

ReviewSegmentation versus integration in visual motion processing

Abstract

Artif. Intell.

Vision Res.

Vision Res.

Vision Res.

Vision Res.

Vision Res.

The Measurement of Visual Motion

Nature

Nature

Foundations of Cyclopean Perception

The Interpretation of Visual Motion

Naturwissenschaften

Biol. Cybern.

Biol. Cybern.

Vision Res.

Science

Perception

Annu. Rev. Neurosci.

Biol. Cybern.

Naturwissenschaften

Biol. Cybern.

Vision Res.

Visual Reconstruction

J. Opt. Soc. Am.

Nature

J. Neurosci.

Review
Segmentation versus integration in visual motion processing