Face Images Intrieval Based On Principal Component Analysis And Maximum Scatter Difference

Face image retrieval technology is so widely used in personal verification, electroniccommerce, financial security and other aspect, it has a great potential application thatbecome one of the most active reseach task for pattern recognition and artificialintelligence. Face image retrieval technology includes face detection technology, facerecognition technology and database technology. The technology of face recognition isthe core of face image retrieval. Over the past half of century, the face recognition hasdeveloped rapidly, only in special environment can achieved the accuracy result of facerecognition. However, a lot of challenges are still leaved to resolve before one canimplement a robust and pratical face recognition application. Among these challenges,the face is sensitive to external environment such as expression, illumination and otherfactor, facial feature extraction is one of the most elementary problems in facerecognition.Face image retrieval in facial feature extraction and recognition were studied in thispaper, the main research work and innovations are as follows:1) Three facial feature extraction methods, which are principal component analysis(PCA), linear discriminant analysis (LDA) and maximum scatter difference, areinstructed. The experiment results on ORL and YALE face databases show that theMSD is better than both PCA and LDA.2) The traditional kernel two-dimensional principal component analysis (K2DPCA)method did not take full advantage of the class information in face images, there areboth “outer class” problem and “hard classifier” problem in face recognition.Therefore,a new face recognition method based on fuzzy kernel two-dimensionalprincipal component analysis(FK2DPCA) is presented.Firstly, it introduces fuzzyconcept into K2DPCA. Secondly, the class separability of criterion will be extended tohigh dimensional feature space by the use of kernel method. And then, selecting theeigenvectors that between-class scatter is greater than within-class scatter afterprojection as optimal projection axis. Finally, it uses the nearest neighbor classifier forface recognition.The experiment results on ORL and YALE face databases show thatthe FK2DPCA is better than traditional methods.3) Considering the so-called “outer classes” and the “inferior” problem in the kernelmaximum scatter difference (KMSD) method, a new method of face recognition based on kernel principal component (KPCA) and fuzzy maximum scatter difference (FMSD)is developed in this chapter. Firstly, the KPCA can be benefit to develop the nonlinearstructures features in faces. Secondly, selecting the eigenvectors that between-classscatter is greater than within-class scatter after projection as optimal projection axis.And then，distribution information of samples is represented with fuzzy membershipdegree in the FMSD. Finally, it uses the nearest neighbor classifier for face recognition.The experiment results on ORL and YALE face databases show that the KFMSD isbetter than KMSD methods.4) Considering the “so-called nolinear” problem and the “outer classes” and “hardclassifier” problem in two-direction maximum scatter difference discriminant analysismethod, a new method (2DKFMSD) of face recognition based on two-dimension kernelprincipal component (K2DPCA) and fuzzy maximum scatter difference (FMSD) isdeveloped in this chapter. Firstly, the K2DPCA can be benefit to develop the nonlinearstructures features in faces. Secondly, selecting the eigenvectors that between-classscatter is greater than within-class scatter after projection as optimal projection axis.And then，distribution information of samples is represented with fuzzy membershipdegree in the FMSD. Finally, it uses the nearest neighbor classifier for face recognition.The experiment results on ORL and YALE face databases show that the2DKFMSD isbetter than other methods.