Study And Implementation Of Face Recognition Algorithm

With the progress of technology, the people pay more attention to the biometric recognition technology. A personal identification system based on face patterns is regarded as one of the most promising biometric recognition technologies because of its convenience, high user acceptance, high visibility, rapid identification, prominent recognition accuracy, difficult counterfeit, low cost and easy extension and application.Firstly, the methods of face recognition are introduced in this paper. The process of face recognition is composed of three parts: face detection and location, face image preprocessing and face recognition which is composed of facial feature extraction, training and matching. Many common methods of face detection, face image preprocessing and face recognition are briefly introduced.Then, a method based on color of face to detect and locate human faces is used. First, a light color image adaptive compensation method is used for image light compensation. Compared with several commonly used color spaces for color extraction, a method about combining YCbCr and KL color spaces is used for color extraction. Then, operations of median filtering, expansion and corrosion are used to the candidate face regions. Finally, face areas are located. Through experiments, we known that such color extraction method has very good effect upon strong and low light of images.Next, preprocessing algorithms of face image are given, which are composed of five parts: image cropping, image conversion, image filtering, image gray transform and size of the normalized image. A method of size of the normalized image is used in the paper: cropping image determined by left and right border of face, location of a pair of eyes and mouth. The standardized method is beneficial to the feature extraction and recognition of face image.Finally, one of face recognition methods based on PCA and wavelet is used. First of all, three of wavelet decomposition to the original image is used. Then, the low weight image and the average weight image are replaced the hominid face image, the problem about big computation of PCA algorithm to be solved. When PCA is used to deal with dimension reduction, a problem which usually be considered is the number of reservations eigenvectors. Another two extraction methods based on PCA algorithm are proposed: Kaiser criterion method and eigenvalue curve method to use in the method of face recognition. Through the experiment, compared with the method of threshold, better results are obtained respectively.