Face Recognition Based On Wavelet Transformation And Kernel Method

Automatic face recognition based on computer is important and popular nowadays.Although some problems are still unsolved a large number of development and achievements have been taken. Mainly the difficulties are extracting efficiently the facial features, boosting both the recognition rate and recognition speed. In this thesis some improved methods are proposed to resolve these problems and achieve good results. A summary of previous achievements in this area is described first. Then a novel angle normalization method is given. According to powerful ability for image operation of wavelet multi-scale transformation, a wavelet coefficient fusing method is proposed. The method combines the projections of vertical and horizontal high frequencies with the low frequencies to make an image vector for recognition so that a good recognition rate can be achieved. A combination kernel function method is also proposed in KPCA using the traditional global kernel—polynomial kernel of strong global feature and local kernel—RBF kernel of strong local feature to make a new combination kernel with good learning and generation. Finally, based on the above method and their improvement, a real-time face recognition system is developed for testing.