Research On Face Detection And Recognition Based On Video

Face recognition uses the inherent characteristics of the human face for authentication, it is different from the ID card which is now being used, and it can improve the security of authentication greatly, and reduce the possibility of forgery and theft. The most important is that it is non-contact identification, and it is motivated. The detection and recognition of face object are the basic framework of face recognition, where the detection of face object is the foundation for the face recognition.In this thesis, the research focuses on the face detection and recognition in video image. The whole research subject consists the following three parts:(1) To detect the face object, the thesis presents a face detection method which based on the skin color and improved Gentle Adaboost algorithm. In this method, it takes4steps to realize the face detection:first, preprocessing the input video frame images; second, segmenting the skin candidate areas in the YCrCb color space; three, taking the facial geometric features to exclude some non-face areas; four, using the way based on the improved Gentle Adaboost algorithm of Haar-like features cascade structure to detect the candidate faces. The experimental results show that the detection method improved the speed and quality of detection, and reduced the rate of error detection that relatives to the single detection way effectively, and it is especially good at detecting the frame which including complex backgrounds.(2) To track the face object, this thesis presents a real-time way based on the previous face object detection results to track the dynamic face target that combining Kalman filter with Camshift algorithm.(3) On the face recognition, this thesis presents a method that using wavelet transform to compress and reduce dimension of video frame image; then using2DPCA algorithm to extract face features; lastly using SVM to classify the face features to get the recognition results. The experimental results show that the combination method can get the effective recognition results and also improve the rate of recognition.The thesis built a human face detection and recognition system in software development platform of VC6.0and OpenCV. Through the detection and recognition experiment for dynamic video, the testing result shows that the rate of face detection reach up to96.0%and the face recognition rate can reach up to96.7%which based on ORL face database.