Research Of Face Recognition Methods Based On Independent Component Analysis

Face recognition technology is one of the hot and difficult topics in computer vision and pattern recognition. Currently, it has an important theoretical value and commercial significance, in recent years, attracts a large number of scholars and research institutions into their research. The goal is to make the computer having excellent recognition capability as human, according to the input facial image to determine the identity of the people. Face recognition process includes:face detection, face segmentation, facial feature extraction, and finally matching and identification, In this dissertation, face recognition methods based on independent component analysis (ICA) are discussed and studied.The main work and innovations are as follows:(1) The dissertation describes the background and significance of face recognition. Introduce briefly the factors which affect the face recognition accuracy and the common face database;(2) Gives a face recognition system framework. Introduce the face detection algorithm commonly and face image preprocessing methods;(3) Feature extraction part:During the face recognition, the features extracted by Independent Component Analysis (ICA) can obtain a good description of the original image, but they can not represent the category information perfectly. Discriminative Common Vector (DCV) is the method which calculates the projection matrix in the zero space of within-class scatter matrix. Compared with the linear discriminate analysis (LDA) method, DCV can get more discriminating features. For this, a new feature extraction algorithm called I-DCV is proposed. Firstly, by using ICA in preprocessed face images, the redundant information in second-order and higher-order can be removed. Then use DCV strike identifying characteristics further. Finally, the Euclidean distance is used for classification. Experimental results show that I-DCV algorithm has a good recognition performance.(4) In order to overcome the one-sidedness and limitations of a single subspace in feature extraction and classification, we propose a face recognition method that extracts features in double complementary space and utilize multi-decision for classification. Firstly, we use ICA and Local Projection Entropy algorithm as the first layer in complementary space to extract global and local features of the face images. Then, for the purpose of solving the problem that features extracted by ICA are lack of classification information, we further extract classification information from the independent features extracted by ICA on the condition that the FLDA and DCV algorithm are used as the second layer in the complementary space. In the classification stage, the test samples are firstly projected to the independent subspace. After that, the samples which are difficult to identify are projected to the LPE space and be reclassified. Finally, the results are been determined comprehensively. Experimental results on FERET and ORL face database show that our proposed method can effectively improve the recognition rate.