Face Recognition Based On Bayesian Method

With the development of information technology, face recognition is becoming a popular research area of biometric person recognition. Face Recognition research area is a multidisciplinary field. It covers image processing, pattern recognition, artificial intelligence, as well as mathematics and so on. So far, people are still researching in some aspects of face recognition such as accuracy, real-time, robust and intelligent to get improvement. In recent years, due to the development of technology and the needs of business, legal profession and other social aspects, face recognition once again become a hot research field.The present face recognition methods mainly are statistical learning models, and in such models, choosing the Bayesian strategy as the classified criterion face recognition has good future. Based on in-depth analysis on Bayesian methods,the main contribution of the dissertation is summarized as follows:1. Haar wavelet is first used to filter the original image. According to the characteristic that The person face is right-and-left symmetry, high and low is asymmetrical, we reserve the vertical high-frequency information of the image. The experiment proved that not only does this may cause the pretreatment computation get big reduction, and not affect too greatly to the precision.2. Using improved Mixed-Gaussian model. That means regarding different classes obedience different normal distribution, and then use this paper's algorithm to get the whole distribution instead of regarding the whole samples obedience a single normal distribution. The experiment proved that this will improve the accuracy of recognition.3. Because the quantity of each class is different, I adjust the prior probability according to the number of samples of each class. It improves the imbalance of the recognition when the sample quantity of each class is inconsistent.