The Primary Research Of Face Recognition

Face recognition in computer analyses the human face images by computer, and extracts the valid information of face images to recognition.Face recognition can be applied widely, for example criminal identification in police, identity verification in driving license and passport, the detection system of bank and Customs and the auto-doorman system. However, the process of reconstruction from 2-D to 3-D is ill-posed. And because of the plenty of expression in human faces, the difference between different ages of one human face, the different illumination environment, Face recognition in computer is very difficult. Face recognition is also involved in some knowledge such as image processing, computer vision, pattern recognition, neural networks and the cognitive process of human beings. So face recognition is a challenging study.This thesis mainly stresses on face recognition and its applications from several aspects:1. Face recognition based on the geometric featureFace recognition based on the geometric feature expresses the human faces by vectors of geometric feature, and identifies the faces by the idea of clustering in pattern recognition. In this part, aiming at the feature of the criminal photos in police, They propose a location method of different apparatus based on local binarization and the curves of point-projection. Then the feature vectors, which have the properties of scale, translation and rotation invariance, are formed. From the similarity compared with the feature vectors in the sample database, the task of face recognition is fulfilled. It is proved highly that this method has advantages of recognition accuracy and robust capability of anti-interference.2. Face recognition based on the deformable templatesThe deformable templates are parameterized simple image models. It can be converged by the energy functions those are defined by our known knowledge and the relevant inhibited condition of the images. They simplify the forms of energy functions based on the known energy function of the deformable templates. They also use genetic algorithms to localthe lowest in the whole. It retains all advantages of the traditional deformable template approach and rectifies some of its weaknesses. Experimental results with some real images show the method has good performance and high speed.3. Face recognition based on the algebraic featureFace recognition based on the algebraic feature expresses the human faces by vectors of feature too. But the vectors are algebraic feature ones. It is the projection of the human faces in the lower dimension sub-space of eigenface. They present a new face recognition method named spectroface in Fourier domain. The proposed method combines the Fourier transform and the wavelet transform for face recognition. And The identification is performance by the low frequency information. The results show that the spectroface is extremely effective for elimination the errors that bring from different expressions and small occlusion.4. Face recognition based on fractalBased on the feature of fractal, They propose a method of face recognition by fractal. The input object image is used as an initial image in a fractal extraction process. The fractal extraction process uses the fractal models of the objects stored in the database, and iterates the initial image foreach fractal model. The difference between the input image and the output of the first iteration for each fractal model is used for identification. This method is also translation-rotation-scale invariance and the effect of identification is good , which is provided by experimentations.