| The early stages of color vision in the primate brain depend upon functionally distinct cells in the retina, thalamus, and primary visual cortex (V1). Within the retina, ganglion cells pool responses from red, green, and blue cones to enhance contrast and discount the illuminant. In the thalamus, single opponent cells differentiate between changes in achromatic and chromatic illumination. In V1, two types of double opponent (DO) cells have been found. The first type of DO cell has been popularized by Livingstone and Hubel. These cells are thought to support simultaneous color contrast, discounting the illuminant, and detecting material boundaries, among other things. The second DO cells, reported by T'so and Gilbert, are thought to integrate chromatic and achromatic vision.; This dissertation defines the DISCOV (DImensionless Shunting COlor Vision) system, which models a cascade of primate color vision cells: retinal ganglion, thalamic single opponent, and two classes of cortical double opponents. A unified model formalism derived from psychophysical axioms produces transparent network dynamics and principled parameter settings. DISCOV fits an array of physiological data for each cell type, and makes testable experimental predictions. Properties of DISCOV model cells are compared with properties of corresponding components in the alternative Neural Fusion model.; The Neural Fusion model has also been used to analyze multispectral and multimodal images in remote sensing and medical imagery. Images are preprocessed to create a multidimensional input vector containing local contrast, color, and texture information at each pixel. An ARTMAP network then classifies input vectors into target and not-target classes.; Benchmark testbeds permit the comparison of DISCOV and Neural Fusion for data fusion. The first testbed is composed of pixels drawn from a MassGIS orthophoto image (0.5 m resolution), taken from an airplane. The second testbed is a NASA Landsat image (15-60 m resolution), taken from a satellite. Both testbed images are views of northeast Boston and its suburbs. The marginal utility of each color vision model cell type is tested to determine its usefulness in image processing and classification. Results varied across testbeds suggesting further study. |
