Human Face Recognition Algorithm Research

As an important application of Artificial Intelligence, automatic machine face recognition is a challenging and difficult task. Its great potential values in both theory and application are always encouraging researchers to commit themselves to the problem. This thesis focuses on the algorithms of face recognition and the main researches are as the follows.The thesis begins with the coarse detection of face, which is the prework of face recognition. Through the study of the feature of human skin, we present a method of skin detection. First we convert the image from the space of RGB to RG in order to weaken the effect of illumination. The Gaussian Mixture Model is set up and E-M algorithm is used to estimate its parameters, which is used to judge weather the pixel is a skin pixel or not. Compared with the traditional skin model, it can change its model parameter according to the different train sample. So this method can fit different skin features.One of the most popular methods used in the face recognition is EigenFace method, which is based on the theory of Principle Component Analysis. It treats the face image as a vector and get EigenValue vector using Karhunen-Loeve convolution. The EigenVectors corresponding to the large EigenValues are similar to the face and are called EigenFace. The linear combinations of these EigenVectors are used to describe, represent and recognize the face image. This paper studies the principle of this method and uses it to extract the features.The theory of statistic learning is wildly used in the filed of pattern recognition. The Support Vector Machine is such a theory to solve two classes problem. Because of its speediness and efficiency, it is wildly used both in research and application. Though the study of the statistic learning theory, especially small train samples, and also the Principle Component Analysis method as referred above, the thesis presents a method of face recognition by EigenFace+ SVM, that is, using EigenFace method to extract feature and using SVM to classify.The paper studies the disadvantage of EigenFace method and learns the advantage of using wavelet to represent the face feature. Gabor function is the only one that can get the lower bound of calculation uncertainty and it can achieve good result both in the spatial and frequency domains. So Gabor function is wildly used in the filed of signal processing. Gabor function has the characteristic of locality,...