Dynamic Face Recognition Algorithm Based On Gabor Wavelet And Svm

As one of the biological recognition technology, face recognition technology is widely used in the identification recognition field, due to its unique stability, friendliness and uniqueness. However, in the field of the dynamic face recognition, on one hand, the face image is easily influenced by illumination, expressions, gestures and other factors,and the impact of these factors will reduce the identification effect of the system, on the other hand, the face recognition system often meets the situation of the new samples to join. If the original samples are only used to learn without considering the useful information of the new samples, it will cause that the recognize performance of the system becomes poor. This paper mainly studies on the two above problems, so, specific research contents are as follows:(1) According to the feature extraction of dynamic face recognition, using the characteristic of the good robustness of Gabor wavelet, this paper proposes a Gabor feature representation method of the dynamic face images based on the block bidirectional statistics. This method first divides the Gabor features into blocks, and these Gabor features are got from the Gabor wavelet transform for face images, and then calculate means and standard deviations of all the rows and columns of each block, and then use all the means and standard deviations to represent Gabor features. The experimental results show that this method not only can effectively extract useful features of dynamic face images, but also can reduce the dimension of the features.(2) According to the new face samples gradually added to the dynamic face recognition system, on the basis of the existing incremental learning methods, the paper proposes an incremental dynamic face recognition algorithm based on the ball ring vector. This algorithm first extracts the ball ring vectors that can become the support vectors from the training samples, and then uses the KKT conditions for stop criteria to realize the incremental learning for the new samples. The algorithm can not only remove the useless samples, but also under the condition of the guarantee for the study accuracy, it can effectively reduce the training time.(3) Combine the Gabor feature representation method of the dynamic face image based on the block bidirectional statistics and the incremental dynamic face recognition algorithm based on the ball ring vector to realize a dynamic face recognition system. Do experiment on the JAN face database and the ORL face database respectively with the algorithm proposed in the paper, and the results show that the system has good robustness. Under the condition of the guarantee for the recognition accuracy, the system has higher learning speed and good dynamic performance for the large face images data.