Accession Number : ADA295126
Title : Computational Methods for Flow Problems - Parallel Algorithms, Flow Control, and Novel Approaches.
Descriptive Note : Final rept. 1 Aug 88-31 Jul 92,
Corporate Author : CALIFORNIA UNIV LOS ANGELES DEPT OF COMPUTER SCIENCE
Personal Author(s) : Gaudiot, ; Von Der Malsburg, Christoph
PDF Url : ADA295126
Report Date : 31 JUL 1992
Pagination or Media Count : 12
Abstract : We have created an object recognition system, in the context of the general goal of contributing to the development of a visual architecture. The system makes use of wavelet transforms, of dynamic link matching, and is of general neural style. We have implemented the system in several versions, as an object-oriented modular program on a workstation, and as a parallel farm structure on an array of transputers. Object recognition from camera images is invariant to translation, scaling and rotation in the image plane, and is robust with respect to lighting and to rotation in depth. We have tested the system on the task of recognizing human faces. With galleries of about 90 faces, the system achieved highly confident recognition on ca. 85% of the input images.
Descriptors : *COMPUTER ARCHITECTURE, *COMPUTATIONAL FLUID DYNAMICS, ALGORITHMS, INPUT, NEURAL NETS, MICROPROCESSORS, CHIPS(ELECTRONICS), PARALLEL PROCESSING, OPTICAL IMAGES, FLOW VISUALIZATION, WORK STATIONS, PATTERN RECOGNITION, SUPERCOMPUTERS, NUMERICAL METHODS AND PROCEDURES, INVARIANCE.
Subject Categories : Computer Programming and Software
Distribution Statement : APPROVED FOR PUBLIC RELEASE