Font Size: a A A

Study Of Face Detection And Recognition Methods In Color Images

Posted on:2008-03-23Degree:MasterType:Thesis
Country:ChinaCandidate:G XingFull Text:PDF
GTID:2178360242972274Subject:Military Intelligence
Abstract/Summary:PDF Full Text Request
Face detection and recognition is one of the active subjects in pattern recognition and computer vision domain, which is the most direct method of identity authentication. It has a wide range of potential applications in video surveillance, identification of certificate and entrance control. Although automatic face recognition is a significant challenge, it's potential value of theory and applications still encourages researchers to make great efforts.This dissertation presents the study of feature extraction and design methods of classifier in color static frontal images. The main points are as follows.1. Study skin region segmentation in color images, a rule function is used to confirm threshold ranges. A rule function is designed according to the character of skin and no-skin pix distributing in the color space. The Cr threshold and Cb threshold is optimized by the rule function, which can include more skin pix and get rid of more no-skin pix.2. An improved face validation method is presented based on SVM arithmetic. A face image is segmented to three parts, and DCT is done for each part. Some DCT coefficients are selected as feature vectors for each part by the energy ratio criterion, which not only containing the main face character but also decreasing the dimension. Three independent SVM classes are trained by these features, and candidate face regions are verified in series structure, which can improve speed of face validation.3. An improved face recognition method based on local KPCA is proposed. Each face image is segmented to several parts according to face image structure. Nonlinear feature is extracted by doing KPCA on the whole face image and each region. All of these features are combined as a whole feature vector to represent the face image, which can overcome shortcomings of kernel principal component analysis (KPCA) method in deal with local image features. This feature vector matrix contains both whole information and local information of the face image, which has well performance.4. In MATLAB platform, an emulator for face detection and recognition in color images is presented, which can do face detection and face recognition in color images, and also can train or recognize gray face image directly.
Keywords/Search Tags:Face Detection, Face Recognition, Feature Extraction, Skin Color Segmentation, Support Vector Machine, Kernel Principal Component Analysis
PDF Full Text Request
Related items