NONLINEAR PROCESSES AT INTERMEDIATE STAGES

描述（改编自申请人摘要）：极度紧张

DESCRIPTION (Adapted from applicant's abstract): A large amount of the nervous system is devoted to vision and we understand little of it as yet. A particular mystery is the function of intermediate levels that are higher than the pointwise processing of the retina but lower than the high levels concerned with objects and scenes (entities about which we are conscious). In the brain, these intermediate processes occupy a grate deal of tissue. The approach in this proposal for studying these intermediate levels is behavioral; it builds on accumulated knowledge about multiple parallel "channels" or "analyzers", especially those sensitive to different ranges of spatial frequency and orientation. Two different kinds of nonlinear processes have been proposed, by us and others, which substantially increase the ability of the multiple channels to explain phenomena of pattern vision. One of the nonlinearities is spatial and can be modeled as "complex channels" which have two stages of linear filtering with a rectification-type nonlinearity in-between. This nonlinearity is needed to explain the processing of "non-Fourier" patterns (i.e. patterns whose spatial Fourier components are not easily related to the observer's performance). The other class of nonlinearity is dependent on the intensive properties of the stimulus and includes both nonlinearities that depend on luminance (e.g. light adaptation in the retina) and on contrast (e.g. contrast-gain controls, which have been proposed for the retina and the cortex). Research is proposed here that would further examine the structure of the complex channels, particularly investigating the relationship between first-stage and second-stage filters' properties and the possible opponency between orientations (and/or spatial frequencies). Research is also proposed into the dynamics of intensive nonlinearities, particularly that due to inter-channel inhibition in the cortex which can be well modeled as a "normalization network". Many of the proposed studies use texture-segregation tasks we have used before --because we have learned enough about them to use them in a rigorous--but other tasks are also proposed both for the sake of generality because they may be better suited for other questions.