| Our perception of a visual stimulus is not only determined by the attributes of the stimulus itself, but also by the context in which the stimulus is presented. Neural correlates of contextual influences on perception have been found in many areas of the brain, i.e., the primary visual cortex (V1), the cortical middle temporal area (Mt), and the optic tectum (TO). In this work, we develop models for neuronal motion processing in order to investigate the possible mechanisms underlying contextual phenomena. We use both analytical and computational tools including computer simulations to compare the results of the models with experimental data. Furthermore, strategies for image motion processing are suggested which might be of potential use in machine vision. The different functional mechanisms that we investigated include synchronization effects in nonlinear dynamical systems, local/global comparisons of visual stimuli, and purely global models for the analysis of visual motion in Fourier space. We find that disorder in the driving forces of an array of coupled nonlinear oscillators leads to transitions from chaotic to regular behavior. This study might help us to understand how synchronization is induced in the brain, since many neurons can be described as stimulus-driven oscillators. We also study descriptive models for visual neurons and show that relative-motion sensitivity is an inherent property of the energy model of complex cells. Based on this observation, strategies for the computation of motion- and orientation contrast are suggested. We further develop an algorithm for motion binding based on temporal filtering in Fourier space. This framework provides the ground for a new model for the retino-tecto-rotundal pathway based on the sparse dendritic structure of wide-field neurons. We demonstrate that global Fourier transformations can be implemented by the tecto-rotundal neuronal population in a straightforward way. As a consequence, it is possible to integrate a Fourier-based mechanism for the computation of structure from motion into the model. Importantly, the spatiotemporal information of the stimulus is retained in the population response at all times. |
