Accession Number : ADA238733

Title :   Analog Neuronal Networks for Early Vision.

Descriptive Note : Final rept.,

Corporate Author : CALIFORNIA INST OF TECH PASADENA

Personal Author(s) : Koch, Christof

Report Date : 1991

Pagination or Media Count : 4

Abstract : The overall goal of the research program funded under this award was to study how a number of early vision tasks, in particular edge detection and motion analysis, can be implemented onto two types of architectures: analog VLSI circuits and the primate's early visual system. In particular, we accomplished the following goals: The introduction of binary or analog line processes, explicitly coding for the presence of a discontinuity in the optical flow, can dramatically improve the final motion flow computed from the time-varying image intensity. The formation of these motion discontinuities are constrained to lie along the location of intensity discontinuities, i.e. edges. We demonstrate this using sequences of video-acquired images. In most formulations of the short-range motion problem, the final optical flow must be regularized, usually by imposing a smoothness condition. Smoothness of the optical flow can explain a number of psychophysical and physiological findings. We implemented the standard Horn and Schunck algorithm onto a model of the magnocellular pathway in the primate's early visual system, in particular onto simple and complex cells in primary visual cortex and onto cells in the middle temporal area (MT).

Descriptors :   ALGORITHMS, ANALOG SYSTEMS, CELLS, CODING, DETECTION, DISCONTINUITIES, EDGES, FLOW, HORN ANTENNAS, IMAGES, INTENSITY, MOTION, NERVE CELLS, NETWORKS, OPTICAL PROPERTIES, PHYSIOLOGY, PRIMATES, SHORT RANGE(DISTANCE), SHORT RANGE(TIME), TIME, VARIATIONS, VISION, VISUAL CORTEX.

Subject Categories : Medicine and Medical Research

Distribution Statement : APPROVED FOR PUBLIC RELEASE