Study And Implementation Of Face Recognition Technology

Human face recognition is attractive in pattern recognition and image processing. Automatic face recognition is an identification technology that analyzes face image with computer to get effective recognition information. It can be applied to security system, human ID, digital surveillance and so on. Face recognition is a biometric technology which possesses great developable potential. Researching on the face recognition technology has great theoretical and practical values.Generally, face recognition consists of three parts: preprocessing, feature extraction and classification. This paper mainly studies on two parts of face recognition: feature extraction and pattern classification.Firstly, this thesis implements and analyzes four typical methods: Principle Component Analysis(PCA), Fisherface, Independent component analysis(ICA) and Gabor wavelets, On one hand, Gabor wavelet exhibits strong characteristics of spatial locality, scale, and orientation selectivity, and the Gabor representations of face images can produce salient local features that are most suitable for face recognition. So it can extract the face feature effectively. On the other hand, ICA would further reduce redundancy and represent independent features explicitly. So this thesis extracts the Independent Gabor features by the Independent Gabor Features(IGF) method, and compares it with Gabor wavelet transform.Secondly, this thesis studies and implements K-NN algorithm about classifier. Furthermore, in order to improve the effect of classification, the paper refers to angle information and the idea of double classification.Thirdly, this thesis proposes a face feature extraction method by 2D Gaussian differential coefficient filters bank which obtains information with different scales and derivatives of face images. Simulation experiment with the features and NN classifier gets a satisfied recognition rate.At last, in order to distinguish the importance of different scales, the thesis calculates the weighted coefficients by using the best square approach principle. The thesis analyzes and compares the results of weighted Gabor with traditional Gabor wavelet transform, where the features are classified by SVM classifier.Simulation experiments prove the validity of these algorithms.