Active-Region Extraction by Learning and Energy Minimization Using Neural Networks

Abstract

In the process of extracting nonuniform target regions such as faces, it is necessary to identify their features. This paper proposes an active-region extraction method based on pattern recognition using a layered neural network. The neural network is used to extract features of previously given sample patterns. During the extraction, the neural network outputs the suitability of the extraction based on the features. Experimental results show that it is possible to extract some occluded parts associated with visible parts of target regions. It is also possible to extract face regions using this method.