Model-Based 3-D recognition System Using Gabor Features and Neural Networks.

摘要

A different approach to pattern recognition was attempted using Gabor features, artificial neural nets, and an image generator. The Gabor features and artificial neural nets are sound biological-based, and the image generator provides complete access to any view of an object. This thesis tested the idea that their integration could form a robust 3-D recognition system. The results of the research showed that the Gabor features together with a neural net were used successfully in classifying objects regardless of their positions, out-of-plane rotations, and to a lesser extent in-plane rotations. The Gabor features were obtained by correlating the image with Gabor filters of varying orientations spaced 15 degrees apart as found in primates' visual systems, and the correlation with each filter was kept separately.