Research On The Methods Of Face Recognition And Eye Location

The recognition of biological features is one kind of automatic technologies which compares the detected physiological features or behavior characteristics with the ones stored in database in advance to identify somebody. In the recognition of biological features are face recognition and eye location which are two hot research topics. This thesis devotes main efforts to the comparision researches of some more common face recognition methods and the automatic eye location research in the stationary images, mainly including:(1) Firstly, Some of face recognition methods based on PCA, 2DPCA and FLDA are comparatively studied. Secondly, the characteristic, application situation and limitations of each method are analyzed. Finally, comparison simulations are performed on these methods and the combinations of these methods by use of AT&T face database, the results show that the method based on 2DPCA has better recognition precision than the one based on PCA does, and that the method based on the combination 2DPCA and FLDA can enhance the classification speed and can have better performance in the constraint of ensuring the recognition precision.(2) With respect to the limitations of the real-time performance and the sensitivity to noises of the eye location algorithm based on the pure topographic features extraction (PTL), a method for fast eye location (GTEL) is proposed by combining Gabor wavelet with the directly extracting topographic features from a human face image in this paper. Firstly, Gabor wavelet transform is used to segment the brow-and-eye region so that the search range for candidate eyes is decreased largely, which can ensure the real time performance of the algorithm. Secondly, the technique of the directly extracting topographic"pit"features from a human face image is employed to locate candidate eyes in the brow-and-eye region, which can enhance the anti-noise performance greatly. Thirdly, symmetry similarity analysis is used to determine the real eye position in the face image. At last, experiments by use of PTL and GTEL algorithms are operated on the images with and without noise respectively, and the results prove the effectiveness of the presented method in this paper.(3) A coarse-to-fine eye location method is presented, based on the combining directly detecting"pit"feature in brow-and-eye region and the eye variance filter. Firstly, with the help of the technology of the direct extraction of"pit"topographic features and wavelet transform, the candidate eyes are searched in brow-and-eye region extracted from the face image by use of Gabor wavelet transform. Secondly, the prior knowledge of face structure and the eye variance filter are employed to determine the real eye positions in the face image. Finally, the comparative experiments are performed on the images with and without noise respectively, and the results show that the proposed method works better in the real-time, accuracy and anti-noise performance than the one purely based on the terrain feature matching does.