Study On Automatic Face Recognition Systems

As the development of the society and the progress of technology, there are increasing demands in automatic verification system. Since some biological characteristics are stable to People and are strongly individual different from one to the others, they can be used as features for verification system. For instance, fingerprint verification system, pupil verification system, speech verification system etc. In comparison with characteristics of above system, the characteristics of face are more directly, friendly and convenient characteristic and can easily be to accepted by the customers verification systemFace is nonrigid and largely variational, so face recognition is a complicated extensive and challenging research task. Recently there are researchful upsurge in the word and significant progresses have been made in the technology of the face recognition. This paper firstly study the background of face automatic recognition system and methods of face recognition, secondly bring forward a method of face recognition based on wavelet transform, principal component analysis and artificial neural network.These are the main contents of this paper:First, accurately seeking-pupil, pretreatment for face, avoiding influence of difference of scale, angle and intensity .Second, using the method of wavelet-transform, extract the stable second-low frequency wavelet coefficient, build face database.Third, for the sake of eliminating the correlations between entities of the image vector and extracting character-vector, which keeps the main class-information and can be used to rebuild original image, the principal component analysis(PCA)was used.Forth, design and mend back propagation (BP) neural network algorithm.Fifth, mended back propagation (BP) algorithm is used in face recognition, and obtains better recognition performance.Comparing with the traditional method, the method proposed here significantly decreases the time of the study and the recognition and remarkable improve recognition ratio. Experimental results presented in thesis verify that the proposed algorithm is effective.