Research Of Face Recognition Based On Color Images

Face recognition is one of the important biometrics techniques which is widely used in many field, including safety inspect of airport and other important place, financial business, intelligent Human-Computer-Interface, etc. It's also broadly used in material management, etc. Compared with other biometric techniques, face recognition has the advantages of being active, secret, continuous, cheap and readily acceptable by the public. Therefore, it's irreplaceable for many applications. Although many face recognition methods have been proposed and some commercial products have been saled, but it's still far away to at large applications. There are many technical problems need to resolve For example, sensation change of the face, the effect of observed orientation and illumination ect. All of that effect recognition rate seriously. Therefore, the research of face recognition is always research hot point of pattern recognition.The thesis has made a series of researching on face recognition-related problems, including automatic face detection, recognition and authentication. As for color image, some novel methods are proposed in feature extraction and face recognition. The main researching includes the following:1, In face detection, the method combining with local feature and apriority knowledge is adopted based on hypothesis-verification framework. The former are used to generating face candidates while the later is used for verification. During hypothesis process, a color image is tansformed into three gray images in R, G, B channel. Base on the result of above step, a novel method for coarse locating eyes and mouth is proposed. The experiment results demonstrated that the method is more robust to poses, illumination, etc especially suitable for detecting rotated faces in image plane. In addition, combining with eye location method based on hybrid projection function, the accurate position of eyes can be located. During verification process, the method on priority knowledge is used to verifying face after face is segmented with the trisection graph method.2, In face recognition, a method is implemented. It use multi-feature to construct hybrid feature, and then SVM classifies these hybrid features. Firstly the input image is transformed three times with wavelet. Secondly, some features are extractec in low frequency image. Eigentface feature is extracted with eigenface method, Gabor features with Gabor filters and local autocorrelation features with the method which is introduced in literature ~[3]. And then, these features are composed to hybrid features. At last, these hybrid features are classified by SVM.The method performes very well in our experiments.