ABSTRACT:

Face recognition has become one of the latest research subjects of pattern recognition and image processing. Although many face recognition techniques have been proposed and many achievements have been obtained, we canpsilat get high recognition rate due to the changes of face expression, location, direction and light. In this paper we study human face recognition based on ensemble techniques. In order to improve diversity ofcomponent classifiers, the idea of bit-plane decomposition is used and moving window classifier is used as a basic individual classifier. The quantized pattern representationspsila layers are used jointly to make a decision. And we mainly study several fused methods which include product, sum, majority vote, max, min and median rules. Experiments results with face images databases show that fusion of multiple classifiers has good classification performance. Moreover, we compare different multiple classifier schemes with other human facerecognition methods.