Face Recognition Based On KPCA And Wavelet Transform

Face recognition has become an issue, which is complex, involving a wide range of applications and broad prospects as for the face' s non-rigid and volatility, and has set off an upsurge and made breakthrough progress in recent years. Though researchers have accumulated rich results in the face recognition technology, they also have encountered some difficulties such as the effective extraction of features, the improvement of recognition rate and velocity of identification and so on. Face recognition is in relation of a lot of technologies, the two keys are extraction of features and classification methods. This paper deals with effective extraction of features of facial characteristics, improvement of the recognition rate and speed on face recognition problems , proposed a feature extraction algorithm based on feature combination, which is proved to be effective. In this paper, the specific contents and innovations include:(1) Introducing and studying the theory involved with face recognition.(2) Complete the face samples image enhancement, the normalized geometric and gray and white processing at the preprocessing time, which would be effective in improving image quality, and lower its computational complexity.(3)Based on the different characteristics of wavelet coefficients: the low-frequency part characterizes the overall image (shape), and the high-frequency part consists of a considerable number of details. Decompose the original tri-tier image by wavelet decomposition, choose the first, second and third floors of the low-frequency band as smooth wavelet characteristics. This will not only retain the overall shape of the face, also weak the local details. At the same time reduces the dimensions of face images, improves the recognition rate and speed.(4) In the area of feature extraction, the KPCA is used in the obtained wavelet characteristics in the paper,to get the three principal components' characteristics in the feature space. The relationship between the characteristics dimension and recognition rate is reserched, and the limitations of traditional methods in feature extraction is pointed out. A new feature extraction algorithm is designed based on the features combination, that is making a partition of primary eigenvector and secondary eigenvector of the obtained three principal components' characteristics, then combine the primary eigenvector with the secondary eigenvector to be the final classify eigenvector of each original sample. At last, input the final classify eigenvectors to the classifier for classification recognition. Experiment shows that the new algorithm is superior to the traditional method of serving one certain type of wavelet as the identifying characteristics. For the more, the recognition speed also has advantages.(5) Build SVM with Polynomial kernel function, and design the multi-class SVM in accordance with the "one-to-one" strategy.Finally, conclude the full work of the text, and point out some contents in need of further study in future.This paper emulates the proposed algorithm by MATLAB simulation, and gives detailed experimental data.