Neural network models for color and brightness perception and binocular rivalry

A neural network model of the early visual system is further developed and analyzed. The model is based on the Boundary Contour System (BCS) for generating an emergent boundary segmentation of a scene, and the Feature Contour System (FCS) for discounting variable illumination and filling-in surface properties of brightness and color within the BCS boundaries. The first part of the thesis applies the model to clarify processes of color and brightness perception. Spatial color and brightness information are represented as spatial patterns of neural activation in hierarchically organized visual processing stages. A model for preserving the invariant contrast-gain and contrast-ratio information in an image at successive processing stages is described. Next, a model for filling-in psychophysically correct brightness increments or decrements is developed, which allows feature flow across boundaries only in the direction of signal increments. Both On-cells that respond to luminance increments and Off-cells that respond to luminance decrements are needed. The temporal dynamics of filling-in are simulated using a backward masking paradigm in which two stimuli are very briefly presented, one after the other. Simulation results match recent psychophysical data collected using this paradigm.; The second part of this thesis simulates quantitative properties of binocular rivalry, the alternation of perception that occurs when the stimuli to the two eyes are sufficiently different that they cannot be fused into a single percept. The simulated model is a reciprocal inhibition oscillator that uses nonlinear feedback signals and slowly habituating chemical transmitters to drive switches between competing populations. Changes in the durations of the dominant and suppressed oscillation phases as a function of differences in stimulus energy to the two eyes are simulated and compared with psychophysical data.