Face Recognition Technology Research Based On The Second Genetation Of Curvelet Transform And Improved Subspace Analysis

In recent years, as improved attention of the international anti-terrorism,security and human-machine interaction of modern society, face recognitiontechnology gets the attention worldwide because of the advantage ofnon-invasive, simple and contactless.The key technology of face recognition algorithm is to extract the stable,unique and differentiable facial features. Recently, the method of transformdomain gets more and more attention, the wavelet transform can only expressthe singular points, which is not the most optimal method to expresstwo-dimensional image with abundant singular curves. The second generationof curvelet transform (SGCT) is sensitive to the singular curves and is widelyused in the field of image fusion and image denoising. Because face imagescontain the abundant singular curves, the paper uses SGCT to extract facialfeatures, and make use of the abundant direction coefficients in order to enhance the robustness in expression changes and illumination changes of theface images.Facial feature vectors extracted by SGCT still stays in the status of highdimension, which rises the computing expense in the process of classifying thevectors, while the sub-space technology gets more and more attention, becauseit occupies small storage space, cost computing expense lowly and can becategorical easily in the field of data dimension reduction techniques. Thepaper uses principal component analysis (PCA), two dimensional principalcomponent analysis (TDPCA), improved TDPCA, and kernel principalcomponent analysis (KPCA) to reduce the facial feature vectors obtainedthrough SGCT in order to reduce the calculation complexity of the subsequentclassifier.In the paper, SGCT is used to extract facial features of face images andsubspace algorithms is used to reduce dimension of facial features, and severalalgorithms are designed. The paper design the algorithm to research therobustness in the expression and illumination changes of the face images.Based on the feature that SGCT extracts multi-scale, multi-directionalcoefficients of the facial features, the paper designs the face recognitionalgorithm to fuse the low-frequency feature vector with the high-frequencyfeature vector with high recognition rate and combined with the improvedsubspace technologies. Based on the feature that SGCT can extract facialfeature vectors with the abundant direction, design the face recognition algorithm to denoise face image and combined with KPCA in order to improvethe face recognition rate. And the paper compares the similarities anddifferences of the several different subspace technologies in the facerecognition. After simulate the designed algorithms on the international facedatabase, the paper analyses the performances of the algorithms. Compareswith the wavelet transform, the designed algorithms show the strong robustnessin the expression changes and illumination changes of the face image, and getshigher face recognition rate and lower time consuming.