Research On Face Feature Extraction Algorithm Based On Gabor Wavelet And CS-LBP

Over past few years, automatic face recognition technique has got a great progress and different kinds of face recognition approaches have been developed. Now, face recognition has a good opportunity to make a further progress and has become one of the most active research areas in computer vision, motivated by the extensive potential application in public security, financial security, human-computer interaction, and so on. Although, automatic face recognition technique achieves a success to some extent, it is still a difficult problem in computer vision. The reason of difficulty is that in the course of obtaining facial image there are some factors to affect the quality of face image, such as variations in pose, facial expression, illumination condition, aging, etc. Therefore, a practical and effective face recognition technique needs to be robust to these variations.Recently, the Gabor filter has been applied to face recognition successfully and has achieved a good performance. In addition to Gabor filter, LBP (Local Binary Pattern, LBP) has also been applied to face recognition successfully. The problem is that the feature dimensionality extracted by combining Gabor and LBP is so high. Therefore, how to reduce feature dimensionality is an urgent thing to solve.Gabor filter and CS-LBP are explored in this paper to develop face recognition algorithm, the reason is to reduce the high dimensionality of extracted feature and to extract the robust feature. The main contribution of this paper is that two face recognition algorithms has been proposed, they are as follows:①A novel approach, combining Gabor filter and CS–LBP for face recognition is proposed in order to reduce the feature dimentionality extracted by Gabor and LBP. In proposed approach, first, we use CS-LBP to replace LBP. Second, we reduce the Gabor feature images through combining the Gabor feature images under different directions and scales. Third, we use CS-LBP to extract features from Gabor feature images after combination. Compared to Gabor + LBP approach, the experimental results on ORL, Yale and FERET of proposed approach show that it reduces the feature dimensionality, storage space and computing time meanwhile gets equal accuracy.②Based on feature fusion thought, multi-level CS-LBP feature fusion approach is proposed for face recognition. We employ CS-LBP to extract more abundant and informative texture features for more times. In this method, first, the CS-LBP is utilized to extract first level features from original face image; then, the second level features are extracted from feature image by CS-LBP again; Likewise we can obtain multi-level texture features and then fuse different levels features to represent original face image. The experimental results on Yale and ORL demonstrate that compared with one level face image features, the approach of multi-level CS-LBP features fusion can improve the face recognition accuracy obviously.