Research On Feature Extraction And Recognition In Face Recognition System

With the development of modern society, the traditional method of identity checking has not been able to meet people's needs. There are increasing demands in new identity checking. Since some biological characteristics are intrinsic and stable, they can be used as the most ideal feature for identity checking. Face recognition have more merits than others. Face recognition technology extract the effective information to carry on the automatic recognition. It involves pattern recognition, computer vision, and physiology. It has been the current research hot spot.The thesis introduces the background and main methods of face recognition. We focus on statistics method and propose a method which bases on wavelet transform and improved neural network of principal component analysis face recognition.First, we eliminate the influence to the recognition in the strength of illumination by histogram and enable the picture to have the uniformity. Then wavelet transform is used to obtain the stable low frequency sub-band of the image in relatively low dimension in order to reduce the complexity of the following algorithm.Secondly, principal component analysis is used to extract the features of the images. The method focuses on the whole gradational relevance of person face image. It extracts feature vectors which retain the main category message in original image space, and reconstructs the original image in the smallest mean error significance. This article uses principal components analysis which has made the improvement to the canonical algorithm.Finally , BP algorithm is introduced to train the neural network for recognition. This algorithm combines the merit of PCA and the adaptability of neural network to improve the recognition rate and the system performance. In the thesis, BP neural network is the improvement algorithm. Firstly the momentum method and the auto-adapted study rate raise the study speed and increase the algorithm reliability. Secondly the change of the outputting level value enhances the system operational speed. This article proposes a method of face recognition which combines the space sorter and BP neural network sorter using their superiority to improve the system recognition rate.Compared with the traditional principal component analysis method, this article combines the merits of many kinds of algorithms. The extracted characteristic has more reflected the difference between the face of person. The method reduces the complexity of the system, and enhances the recognition rate and the system performance. This article uses Cambridge ORL face storehouse to carry on the experiment and to validate the system performance.