Face Recognition Based On Bayesian Decision Fusion Of The Algorithm

The technique of face recognition is an application of the image manipulation, it has very strong aptitude, the application of it is feasible, it brings convenience and safety. Face recognition is a focus of the research in the field of Applied Mathematics, Pattern Recognition and Computer Vision, it provides a kind of high reliability, good stability approach of identity appreciation. This technology involves many related disciplines, and the key technology is the Feature Extraction and Classification Method.This paper develops study based on these emphases. These contents are as follows:The preprocessing phase uses exponential attenuation to complete illumination compensation of the original images and uses the wavelet decomposition of the processing images. Low frequency images are still steady under the changes of face expression while high frequency images reflects the face of the detail feature, its role can not be ignored, therefore I present a wavelet sub-graph fusion method.This method can decrease computational complexity and improve the computational speed under the precondition of guarantee the recognition effect.In the feature extract phase, we not only detail the theory and the specific algorithm of the principal component analysis and linear discriminate analysis, but also extract the two-dimensional feature face feature, the two-dimensional linear discriminate feature and the feature face feature of image differences.Different classifiers have their particular advantages. So combining different classifiers will be a wise choice. This paper discusses the accuracy of bayesian fusion, and then gives two methods of classification of multi-channel bayesian fusion based on posterior probability: one is a weighted data fusion, and the other is hierarchical bayesian fusion classification. Through weighted fusion of posterior probability of two features (pca, 1da) and through hierarchical bayesian fusion of image differences'feature face feature and linear discriminate feature. I make three groups of experiments. Theoretical analysis and experiments show that the recognition accuracy of multi-channel bayesian fusion system has better performance than any single channel recognition accuracy in ideal circumstances, and obtains satisfying recognition effect.