Research And Realization Of Robust Face Recognition Algorithm

The purpose of biometrics is to make use of people's personal features to authenticate the identity. Face recognition technology is a natural and direct biometrics method. Compared with other biometrics method such as fingerprint, iris, retina, and gene etc, face recognition system is direct, friendly, convenient, and more inclined to be accepted by users. Furthermore, mental information hard to obtain by other recognizing systems can be gained through analysis of expression and gesture. It is a key technology of man-machine interaction and intelligent living environment in the future. It has great application value in economy, security, social insurance, criminal investigation, and military fields.Face recognition technology involves psychology, physiology, AI, pattern recognition, computer vision, image analysis and processing fields. Especially, it's basic revealing of human intelligence. It's the most typical and difficult pattern recognition. The research and solving of this problem do benefits other object recognition problems. So face recognition of great theoretical research value becomes one of the important subjects of such basic research.The focus of work is the research and realization of robust face recognition algorithm. The original work including:1. Proposed a new digital filter for preprocessing of face image. Firstly giving the model of the filter, then discussing about quality of the filter and its applicationto feature strengthening. Finally discussing its application in preprocessing of face image combined with PCA.2. Proposed a new structure of committee machine and a new method of output fusion. First represent a common problem in classification i.e."bias/variance" dilemma, which is the basis of committee machine design. Then use RPROP algorithm to realize MLP, while OLVQ3 to realize RBF. Feature extraction from hidden layer of MLP is analyzed. These two ANNs would be used as experts in committee machine. Last comes committee machine principle and its realization. Through comparing methods and results of various references, and analyzing their conclusions, committee machine is proposed. The theory and experiment both show that it could retain a high rate of recognition steadily. It is a robust algorithm for face recognition.3. Analyzing two feature extracting methods widely used, i.e. PCA for feature extract and ICA for feature extract, showing its specific algorithm and experiment result. Performing experiment result analysis. 10 algorithms are realized in this paper. Based on ORL face database, the result of such algorithms on different amount of samples is represented. Through comparison, analysis and discussion are furthered.