The Study Of Face Recogniton Methods Under Heterogeneous Light Sources

Face recognition technology under the visible light has made great progress since the nineties,but its vulnerability to changes in ambient lighting effects led to a sharp decline in the recognition rate.To some extent,using near infrared camera solves the problem of adverse effects of ambient light on face recognition.However,the use of near infrared camera in the face recognition system will result in face recognition under heterogeneous light sources,including face recognition under the same light sources and face recognition under different cross homogeneous light sources.Among them,the face recognition under different cross homogeneous light sources refers to two-way cross-recognition problem,including the near infrared face image registration-the visible face image authentication and the visible face image registration-the near infrared face image authentication.Due to the visible and near infrared image are in different imaging modalities,there are significant differences between the visible face and the near infrared face of the same person,and the small number of existing face recognition methods under different cross homogeneous light sources mostly involves only the visible face image registration-the near infrared face image authentication in one direction,therefore,when the near infrared imaging technology to be adopted to solve the ambient illumination problem of face recognition,face recognition under heterogeneous light sources is still an urgent problem.Contents of this subject covers the entire process of face recognition under heterogeneous light sources,including face detection,automatical determination of different light types,facial feature extraction and face authentication.Specific studies are as follows:(a)Face detection under heterogeneous light sources,applys classifier trained by adaboost algorithm to solve the problem of face detection under heterogeneous light sources,and gets two face images of different sizes cut from one human face image at the same time based on the result of face detection according to the need of this subject's face recognition;(b)Automatical determination of different light types,the method was proposed thatautomatically determine different light types based on gray distribution of the full video frame image,and experiments showed the feasibility of this method;(c)Facial feature extraction under heterogeneous light sources,firstly introduced the method based on statistical characteristics and the method based on structural feature points,and on the basis of facial structure feature points proposed three structural features based on the distances among feature points,combined with the nearest neighbor classification algorithm compared and analysed the face recognition results of a variety of feature extraction methods on the visible-near infrared face library established in this study.Ultimately came to a conclusion that the fusion of 2DPCA and the proportion of distance among feature points is superior to other methods;(d)Face authentication under heterogeneous light sources,analyzed a variety of situations that may be occured when using near infrared camera in the actual face recognition system,according to the actual circumstances,designed two kinds of face authentication schemes,including face recognition under the same light sources and face recognition under different cross homogeneous light sources.Among them,2DCPA feature extraction method combined with the nearest neighbor classification algorithm was used in the face recognition under the same light sources;the fusion of 2DPCA and the proportion of distance among of feature points feature extraction method combined with the nearest neighbor classification algorithm was used in face recognition under different cross homogeneous light sources.Finally,set up a face recognition system under heterogeneous light sources with VC++ 6.0and OpenCV platform,using near infrared imaging technology to solve the adverse effects of the existing ambient light in the visible light face recognition,and using the face recognition method under different cross homogeneous light sources raised by this subject to solve two-way cross-recognition problems about the near infrared face image registration-the visible face image authentication and the visible face image registration-the near infrared face image authentication.