Research On Face Recognition Algorithm Based On Texture Feature Fusion And SVM

Face recognition technology is anti-counterfeiting, user- friendly and recognition active and it not only plays an important role in security, intelligent video surveillance, etc, but also shows great practicability in the mobile payment of e-commerce and so on. In practical applications, face recognition technology has to overcome the interference brought by illumination, the variation of gestures, shade and the variation of expression. Specifically, the simultaneous emergence of light and posture change will serious ly affect the effect of face recognition. C urrently, the interference problem incurred by illumination and posture change is still not completely solved. So the research on how to overcome the illumination and posture change is of great significance to the interference of face recognition.Lots of experts and scholars have carried out a series of research to solve the interference problem of face recognition. Among those algorithms, the face recognition algorithm based on texture feature attracts more and more attention. This paper focuses on facial texture feature extraction and fusion algorithms, and improves the robustness for the illumination and posture change of the face recognition system. The specific works are as the following:① This paper mainly researches the Gabor Wavelet Transform, Non Subsampled Contourlet Transform(NSCT) and Local Binary Pattern(LBP) texture feature extraction algorithms. Through the experiment and analysis of the realization of facial feature extraction algorithm, the Gabor wavelet could extract multi-scale 、multi-direction facial texture feature. However, the face feature has the problem of high frequency information loss and large dimension. The high frequency characteristics extracted by NSCT transform is robust to illumination, but failed to make full use of the low frequency subband information in face feature. The LBP operator is able to express facial neighborhood texture feature, but vulnerable to light and face expressions. In order to fully combine the advantages of LBP operator and NSCT, this paper puts forward a method using uniform LBP operator to extract the NSCT transform multi-scale 、 multi-direction high frequency subband texture feature. Then, doing analysis of each uniform LBP feature of high frequency subbandis and combining the statistical information together. Also, a new texture feature(Histogram of Uniform Local NSCT Binary Pattern, ULNBH) is obtained.This feature achieves the scale information expression and dimension reduction.② Although the ULNBH characteristic combine the advantages of LBP operator and NSCT,But it still lacks of low frequency information, this paper combined the proposed ULNBH characteristics and the Gabor features in the feature layer, so as to get a fusion characteristics with more complete facial texture feature information(G-ULNBH characteristic). In order to further verify the effectiveness of G-ULN BH feature, this paper selects Support Vector Machine(SVM) as the classifier to recognize the face using G-ULNBH feature. First, the high-dimensional feature vectoris reduced using Principal Component Analysis(PCA) methods, and then the fusion characterist ic after reduction is identified by the SVM. The O RL, Yale and self-built face libraries are adopted in the simulations and the recognition effect on the face features extracted by some other algorithms are analyzed comparatively. Through the experiments, the effect of the G-LUNBH+PCA+SVM recognition algorithm is proved to be the best, which shows the fusion characteristicis more robust to illumination and posture variation.