Face Recognition Algorithm Research

Face recognition is a computer analysis of the characteristics of face images in order to achieve identity verification technology, because of its burden of proof in court, cardholder identification, video surveillance and other aspects of the application of great value, is currently subject to governments and their military, security, intelligence agencies, as well as widespread concern in scientific research and attention. A complete face recognition system consists of face detection, face image preprocessing, face feature extraction, image features such as inter-similarity calculation module. This study includes the following aspects:(1) Face image preprocessingSome of the basic pre-processing methods are proposed in this paper, such as: histogram equalization method of image enhancement, image restoration, as well as image rotation and pure face partition. Experiments show that these methods can greatly enhance the recognition rate, and save a lot of storage space. Through these methods, we can get the best recognition results of image(2) Application of wavelet transform in face recognitionBecause of their good time-frequency localization properties, wavelet transform can provide the most substantial features of human face and weaken the interference noise to reduce the computation, it will be used in face recognition wavelet image compression and image fusion with good prospects for development.(3) Study and compared four typical methods of facial feature extractionMethod of principal component analysis (PCA), principal component analysis based the Eigen face method, the face image region as a random vector. Obtained by using K-L transform a linear combination of orthogonal base to describe facial expression and close images, and extract facial features.Fisher linear discriminant analysis method is to model the original high-dimensional samples of the optimal discriminant projection vector space in order to achieve classification of the information collected and compressed feature space dimension effect, experiments show that Fisher linear discriminant analysis method obtains better results than the traditional PCA.Discrete cosine transform (DCT) is a good method of data compression. This paper introduces the DCT in the lower-dimensional face recognition applications: DCT transform coefficients of the finite matrix coefficient of the upper-left corner contains the majority of face image information, which is more conducive to identify some of the information, so only part of the retention of this factor can to the purpose of dimensionality reduction.Independent component analysis (ICA) is based on the signal characteristics of higher-order statistics analysis. Of the proposed algorithm, this article in the ORL face database and Yale carried out on a large number of repeated experiments, the experimental results show that the method of independent component analysis has the highest recognition rate, and need the largest calculations. The method of Principal Component Analysis to has the lowest rates of face and need to minimize the amount of calculation.(4) In the classifier design, this study and compared the K-neighbors (K-NN) classification method and support vector machine. Among them, the support vector machine method has the global optimum, simple structure and strong ability to promote the advantages, so as a structural risk minimization to achieve the specific criteria, has been extensive research and development in recent years. Finally, this paper introduces a binary-based template matching method for face recognition. This method through calculates two overlap front picture prime number the proportion which occupies in its total front picture prime number to weigh two person face image the similarity. In Yale face image Database, the experiments show that the method can be faster than the speed of the PCA method to identify a higher rate, binary template matching method can obtain a higher robustness of the light and robust expression in order to more practical applications.