Research And Application Of Face Recognition Algorithm Based On Bayesian Theory

As a convenient and easy collected identity verification technology,face recognition owns a wide applications and markets.It has a great significance to make life convenient and ensure information safety.In our work,we main make research on face recognition based on Bayes theory.The basic idea of face recognition based on Bayes theory is to estimate the probability distributions of the training samples from different categories of face images,to calculate the probability of a test sample by Bayes theory,and to make classification by the probability.In this paper,we make focus on two Bayesian face recognition frameworks: naive Bayesian and joint Bayesian,after analyzing the weak points of those algorithms,we optimize the original method,what’s more,we propose an optimized joint Bayesian method and apply it to the face recognition system.We propose an optimized naive Bayesian method.Its basic idea is as follows.The method estimates the probability distribution of each pixel at each gray level in the face images,then it performs Laplace smoothing to resolve the zero probability problem.Because the pixels in the image is dependent and the naive Bayesian method classifies the test sample into the class with maximum conditional probability,the maximum filtering is used to optimize the probability distribution matrix to resolve the probability.Finally we use the naive Bayesian framework to conduct classification.Experiments on some face databases show that the proposed algorithm is effective.We propose a weighted sub-similarity joint Bayesian algorithm whose main steps of the algorithm are as follows: after defining two novel similarities for samples based on the similarity measure of joint Bayesian face recognition method,we fuse these two similarities using different weights to produce a new decision similarity.The weights are obtained by the logistic regression method.Experiments on some face databases show that the proposed algorithm is effective.In addition,we apply the joint Bayesian method into our face recognition system,in which,a convolutional neural network is used for feature extraction of face images.We introduce the self-similarity to solve the problem that the joint Bayesian method does not perform well in real scenes.We collect two face images database from real scene.Then a lot of experiments show that the method optimized by self-similarity has a better accuracy when compared with original method,and it also has robustness to light.