The Applications Of DWT And FAST PCA And SVM In Human Face Identification

Currently, Automatic system of face recognition is a research focus in the field of pattern recognition, computer vision, artificial intelligence and so on. The information of human face could be gained from non-contact method, such as cameras, so it is very suitable for identification authentication. In the modern and rapidly developing society, a new and reliable type of security system is urgently needed. This system has to protect the safety of people’s properties and lives. Meanwhile, it should not affect people’s daily life. The system of face recognition could meet the requirements of people perfectly. Therefore, it is widely used.A system is achieved by using the algorithm of face recognition, which combines with DWT (Discrete Wavelet Transform) and FAST PCA (Fast Principal Component Analysis) to extract features, and SVM (Support Vector Machine) based on the algorithm of improved binary tree, is used as a classifier. At last a prototype system is given in this paper. The system of face recognition used the improved algorithm is tested in the MATLAB7.10.0(R2010a)software with the ORL and YALE database, and the results show that not only the recognition rate but also the efficiency are heightened, which compared with the other algorithms by the simulation test. This paper contains the main contents as follows:(1) When the images are captured by CMOS, the face images are influenced by the objective factors such as noise, light, location. So the images need to be preprocessed to eliminate the influence on face recognition system, which includes image denoising, gray-scale processing, geometric normalization, illumination compensation and so on. Then the algorithms of face detection are introduced in this paper, the approximate location of face in the image could be exactly located by the face detection algorithms.(2) The problem of traditional PCA is lead to the large dimension of scatter matrix of samples when making images converted from two-dimensional matrix into one-dimensional vector, which will increase calculating complexities of eigenvalue and eigenvector. Moreover, the operation time of feature extraction is longer than the other algorithm when using PCA for feature extraction. Therefore, a modified PCA is put forward in this paper. In the stage of feature extraction, discrete wavelet decomposition is used to extract the sub-image of low frequency, then a modified PCAis used to dimensionality reduction, the test demonstrates that the algorithm of feature extraction, combined with DWT and FAST PCA, reduced the time of feature extraction greatly.(3) The stage of classification is focused on the basic principle of SVM and SVM is extended to deal with multi-class classification problem by the algorithm of improved binary tree. Because of SVM’s unique advantages in solving the problem of non-linear, high dimension and small sample, so the SVM based on the algorithm of improved binary tree is used as a classifier is proposed in this paper. The results of experiment show that the face recognition algorithm is proposed which could acquire a high recognition rate by the simulation test in the MATLAB software with the ORL and YALE database.(4) A face recognition system is implemented by MATLAB which uses the algorithm and proposed in this paper. To some extent, this system of face recognition meets the basic demands of security, so it has a certain practical value and broad prospect of application.