The Design And Implementation Of Face Recognition System Based On Statistic Features

The Face Recognition System generally works as follows:it starts with detecting the face region from video sequences or a static image, and then extracts facial features in the obtained region. The face image is watched in comparing to the face model in the database through the face recognition algorithm. The performance of face recognition system is affected by the factors such as facial feature extraction method, face modeling and classification method. In this thesis, it designed and implemented a face recognition systems to video sequences based on statistical feature extraction and Hidden Markov Modeling approach. In particular, face image would be blocked to discrete cosine transform DCT at first, and then the transform coefficient matrix could be done principal component analysis PCA to obtain the feature vectors which would be used as the observation vectors in the training of Hidden Markov Model HMM. Using this recognition method has the following advantages:firstly, the feature extraction method is to reduce redundancy in order to improve the image feature extraction speed. Secondly, the hidden Markov model HMM which models the human face accuratly could ensure the identification accuracy. Thirdly, HMM face recognition method is robust to the changes of light, gesture, blocking material due to using the two-dimensional statistical features of the face. Finally, the face database is easy to maintain which the faces correspond with the sample model so that the Add/Remove operation is easier.This thesis researched and designed a face recognition system.At first he face images should be collected and trained to create a face database at first. The database is used in identifying individuals to match face images. We enter a video sequences with the static background, and extract the movement human body using the frame difference method. And then we detect the face region by the Adaboost algorithm, and do a series of preprocessing operations, including filtering noise removal, binarization, to this area. Block the face image to discrete cosine transform DCT at first, and then do principal component analysis PCA to the transform coefficient matrix to obtain the feature vectors. Namely, human facial feature extraction. The extracted features are used as the observation sequence of Hidden Markov Model HMM, and the observation sequence is matched with the face HMM model in the library, that is recognition. So a face recognition system based on video sequences is achieved. ORL and face database established are uesd to test the recognition module of the system, that is, feature extraction in DCT and DCT+PCA are contrasted at the input of static imge. Then we input the video sequences to find the application effect of the system in the case of established database. Experiments show that the method could effectively segment the human body from the video sequences and detect the face region to some extent. It would recognize the faces when the final application used the same camera with which used in the face database acquisition, or the resolution is higher than the resolution of the sample collection. On the one hand the system realized face recognition in the the laboratory initially, on the other hand it would provide application base for the Human-Machine Intelligence interaction, video surveillance and identity authentication in the long run.