Occluded object recognition by Hopfield networks

摘要

A new method to use a Hopfield neural network for objectnrecognition is proposed. Object recognition is treated as a subgraphnmatching. A system consisting of one global network and severalnsub-networks is constructed. The sub-networks are dynamically changednand the outputs of the global network and the sub-networks are fedbacknto each other to complete the subgraph matching. This method avoids thenlocal minimum problem arising from the use of one single Hopfieldnnetwork and it also uses much less time than the simulated annealingnalgorithm. Computer simulation shows it can efficiently recognizenobjects in occlusion