Disparity tuning as simulated by a neural net

Adaptive Neuromorphic Circuit for Stereoscopic Disparity Using Ocular Dominance Map

Stereopsis or depth perception is a critical aspect of information processing in the brain and is computed from the positional shift or disparity between the images seen by the two eyes. Various algorithms and their hardware implementation that compute disparity in real time have been proposed; however, most of them compute disparity through complex mathematical calculations that are difficult to realize in hardware and are biologically unrealistic. The brain presumably uses simpler methods to extract depth information from the environment and hence newer methodologies that could perform stereopsis with brain like elegance need to be explored. This paper proposes an innovative aVLSI design that leverages the columnar organization of ocular dominance in the brain and uses time-staggered Winner Take All (ts-WTA) to adaptively create disparity tuned cells. Physiological findings support the presence of disparity cells in the visual cortex and show that these cells surface as a result of binocular stimulation received after birth. Therefore, creating in hardware cells that can learn different disparities with experience not only is novel but also is biologically more realistic. These disparity cells, when allowed to interact diffusively on a larger scale, can be used to adaptively create stable topological disparity maps in silicon.

Spatial frequency integration for binocular correspondence in macaque area V4

Neurons in the primary visual cortex (V1) detect binocular disparity by computing the local disparity energy of stereo images. The representation of binocular disparity in V1 contradicts the global correspondence when the image is binocularly anticorrelated. To solve the stereo correspondence problem, this rudimentary representation of stereoscopic depth needs to be further processed in the extrastriate cortex. Integrating signals over multiple spatial frequency channels is one possible mechanism supported by theoretical and psychophysical studies. We examined selectivities of single V4 neurons for both binocular disparity and spatial frequency in two awake, fixating monkeys. Disparity tuning was examined with a binocularly correlated random-dot stereogram (RDS) as well as its anticorrelated counterpart, whereas spatial frequency tuning was examined with a sine wave grating or a narrowband noise. Neurons with broader spatial frequency tuning exhibited more attenuated disparity tuning for the anticorrelated RDS. Additional rectification at the output of the energy model does not likely account for this attenuation because the degree of attenuation does not differ among the various types of disparity-tuned neurons. The results suggest that disparity energy signals are integrated across spatial frequency channels for generating a representation of stereoscopic depth in V4.

A threshold explains modulation of neural responses to opposite-contrast stereograms

Disparity-sensitive neurons respond to contrast-inverted stereograms (aRDS) that do not evoke depth percepts. This is in conflict with the idea that such neurons are the direct correlate of depth perception. However, the output of neurons responding to aRDS may be further processed: neurons at later processing stages show weaker responses to aRDS than early stage neurons. Here, we show that such a response hierarchy emerges in a three-layered neural network. A numerical analysis demonstrates that threshold operations can largely explain the network's behavior as well as the electrophysiological data. An extension of the energy neuron model for disparity-sensitive neurons predicts increased responses to aRDS for an identifiable sub-class of cells and can thus be tested in electrophysiological experiments.

Hierarchical processing of horizontal disparity information in the visual forebrain of behaving owls

According to their restricted receptive fields and input-filter characteristics, disparity-sensitive neurons at early processing levels of the visual system perform rather ambiguous computations; they respond vigorously to disparity in false-matched images and show multiple response peaks in their disparity-tuning profiles. On the other hand, the perception of depth from binocular disparity is reliable, thus raising the question as to where and how in the brain additional processing is accomplished leading toward behaviorally relevant disparity detection. To address this issue, tuning data during stimulation with correlated and anticorrelated random-dot stereograms (a-RDS) were obtained from 52 disparity-sensitive visual Wulst neurons in three behaving owls. From the disparity-tuning curves, several quantitative measures were derived that allowed to determine the response ambiguity of a cell. A systematic decline of response ambiguities with increasing response latencies was observed. An increase in response latencies of neurons was correlated with a decrease of the strength of responses to a-RDS. Declining responses to a-RDS are expected for global detectors, because an owl was not able to discriminate depth in psychophysical tests with a-RDS. In addition, suppression of response side peaks was increased and disparity tuning was enhanced with growing response latencies. These results suggest a functional hierarchy of disparity processing in the owl's forebrain, leading from spatial filters to more global disparity detectors that may be able to solve the correspondence problem. Nonlinear threshold operations and inhibition are proposed as candidate mechanisms to resolve coding ambiguities.