Accession Number : ADA296381
Title : Design for Silicon Cortex.
Descriptive Note : Final rept. 1 Oct 94-29 May 95,
Corporate Author : TRANSITIONS RESEARCH CORP DANBURY CT
Personal Author(s) : Weiman, Carl F. ; Laine, Andrew F.
PDF Url : ADA296381
Report Date : 07 JUL 1995
Pagination or Media Count : 56
Abstract : This research analyzed mathematical properties of the discrete Gabor filter and tested it on real imagery. Several unexpected results emerged. The discrete Gabor filter, is not a wavelet, and is not a complete image decomposition function. To complete the set, additional filters are required. The Gab or filter measured binocular disparity to 1/100th pixel accuracy. However, there was an large phase spread across contrast boundaries, due to the poor bandpass of analog video signals feeding the frame buffer. In recognizing handwritten characters, the discrete Gab or filter performed more poorly than wavelets, because of the lack of components which would have recognized junction topologies of characters with 'X' and 'Y' sub-geometries. This failure is related to the lack of completeness discovered in the mathematical analysis. Such negative results preclude the design of a general purpose silicon cortex architecture. Nevertheless, later studies could augment the components proposed here, testing them on a high performance general purpose processor such as the C80 system from GIC, or the CNAPS. (KAR) P. 1
Descriptors : *IMAGE PROCESSING, *MATHEMATICAL FILTERS, BUFFERS, CONTRAST, ACCURACY, PARALLEL PROCESSING, BOUNDARIES, MATHEMATICAL ANALYSIS, SILICON, IMAGES, FRAMES, VIDEO SIGNALS, FEEDING, DECOMPOSITION, ANALOG SIGNALS, VISUAL CORTEX.
Subject Categories : Numerical Mathematics
Distribution Statement : APPROVED FOR PUBLIC RELEASE