A Research On Face Recognition Technique Based On Shearlet Transform

Face recognition technique is an important component of biometric recognition techniques. After facial parts being segmented from an image, some discriminative facial features are extracted from facial parts of image, by these discriminative features, classificatory identification (such as face identification in applications of entrance control system) and recognizing classification (such as face recognition in applications of personal identity system) are then conducted. Facial features, likewise other biometric features, generally can be considered as one’s inherent characteristics once he/she being born. These biometric features are discriminative between any two persons, because they are unique among people and nearly unchangeable over some extended periods. Comparing with other biometric recognition methods, the most important advantage of face recognition is the untouched and non-invasive. Face recognition systems are inclined to be widely accepted in various sectors, the main cause may be the face image can be collected by an untouched and a relative distant mounted device, in addition, the recognition system can be operated in a non-intervention and non-notice manner. Face recognition techniques have been broadly applied in many sectors such as special region’s security control criminal investigation, customs access control, personalized business services.Principal component analysis (PCA) is the first pattern recognition method being applied in face recognition. The 2-dimension PCA (2DPCA) is an improved version of PCA when being utilized in face recognition. By PCA, one 2-D image is firstly re-arranged as one-dimension(1-D) data series by row-by-row or column-by- column, and PCAs are then calculated from the 1-D data series. And by 2DPCA, the covariance matrix of an image is calculated directly from the image matrix, so the rank of covariance is significantly small, and less calculation is needed. The thesis firstly reiterates the principles of these two recognition methods, and conducts corresponding experimentations of face recognition the two methods. Due to support vector machine (SVM) having been proven very suitable to be applied in small sample-size and high-dimension features recognition problem, with excellent generalization capability in the many pattern recognition applications, this thesis secondly conducts the simulation experiments and analysis on the selection of different kernel-functions, the results of these experiments show that the radial basis function (RBF) is the best choice among other kernel functions for SVM construction.Thirdly and also the key-point research being conducted in this thesis, is the shearlet features extraction from face image and face recognition based on shearlet transform. The shearlet features of a face ares actually the shearlet (a special wavelet) coefficients when the face image is transformed by shearlet transform. Shearlet is a multi-scale analysis tool, a special wavelet based on classic wavelet, with the capability of multi-resolution analysis and multi-scale directional representation, and with a better sparse performance when comparing to contourlets and other typical wavelets. This thesis reconstructs and analyzes high and low frequency domain of discrete shearlet transform of image, and then extracts shearlet feature from face image, wherein coordinately utilizing radial basis function (RBF) neural network. Finally, a SVM classifier is constructed by RBF kernel function, and face recognition experiments are carried out on the ORL face database and Yale face database. The results shows that:face’s shearlet features have excellent recognition performance in these experiments, and shearlet features are considerable choice when building face identification or face recognition systems..