Face recognition is an important part and a hot research topicof computer vision, pattern recognition and image processing,andit is also an ideal basis and the most natural approach of identityidentification.This dissertation analyzes the physiological essences anddifficulties in the process of face recognition in depth, and on thebasis of scientific classification, it makes a detailed discussion ofthe design principles, application characteristics, and existingproblems of the main face recognition algorithms on the basis ofscientific classification. Based on these, some new methods arebrought forward, which are detailed as follows: 1. Kernel principal component analysis (KPCA) is applied toface recognition, so that nonlinear classification of faces isimplemented by nonlinearly classifying kernel principalcomponent vectors. This method appears to have high ability todistinguish whether a face belongs to the database or not, as wellas high right ratio of the first match and that of the first tenmatches.2. In the implementation of face recognition based onindependent component analysis (ICA), the right recognition ratioand the right ratio of the first match and that of the first tenmatches have been obviously improved by using Kernel principalcomponent vectors as inputs instead of principal componentvectors.3. An "independent multidimensional component analysis(IMCA)" is originally brought up, and a consequent algorithm isacquired on the basis of Guassian kernel function. IMCA is atheoretical extension and generation of ICA. In the experiment offace recognition by using IMCA, the extremely high right ratio ofthe first match and that of the first ten matches are observed.4. It is put forward and verified that the combination of KPCAand IMCA is a possible approach to realize face recognition inhigh performance, in which distinguishing database-in faces fromdatabase-out faces and determining the identity of faces are basedon KPCA and IMCA respectively. |