A Study On The Methods Of Face Recognition Based On Optimized LDA And RBF Neural Network

Comparing with other technologies of biometric recognition, the technology of facerecognition has so many advantages such as non-intrusive, initiative, convenience and friendly.Therefore, it has more wider application. The technology of face recognition has become a hotresearch focus in many subjects such as artificial intelligence, pattern recognition.In this paper, the general situation about the development of face recognition technology isfirstly introduced. Then the problems of feature extraction and classification are discussed.The second chapter of this paper presents an optimized LDA algorithm to overcomequestions existing in the traditional LDA algorithm for FR in this paper. The between-classscatter matrix is redefined in order to make the traditional Fisher criterion optimal and eliminatethe effect that the edge of class has on selecting the optimal projection. At the same time, itavoids computing the inverse of matrix and solves the Small Sample Size (SSS) problem bymeans of factorization. So we can choose the appropriate value of e by adopting experientialmethod. And then the optimal effect of face recognition is got. Experimental results show thatthe recognition rate of this method is superior to the traditional LDA.The third chapter of thispaper presents a new method of face recognition based on local singular value features and radialbasis function neural network. I use several small windows to partition the face image, thencalculate the singular value features of each window and rearrange the singular value featuresinto a new single singular value features vector according to the lexicographical order. In theRBF neural network, I set the number of hidden units by simple clustering and adjust theparameters of hidden units by adopting a hybrid learning algorithm, which combines the gradientparadigm and the linear least square(LLS)paradigm. Experimental results show that thisrecognition method I put forward is a superior method because it has higher recognition rate,more rapid recognition speed and can be realized simply.