Multiresolution Feature Based Fractional Power Polynomial Kernel Fisher Discriminant Model for Face Recognition

This paper presents a technique for face recognition which uses wavelet transform to derive desirable facial features. Three level decompositions are used to form the pyramidal multiresolution features to cope with the variations due to illumination and facial expression changes. The fractional power polynomial kernel maps the input data into an implicit feature space with a nonlinear mapping. Being linear in the feature space, but nonlinear in the input space, kernel is capable of deriving low dimensional features that incorporate higher order statistic. The Linear Discriminant Analysis is applied to kernel mapped multiresolution featured data. The effectiveness of this Wavelet Kernel Fisher Classifier algorithm is compared with the different existing popular algorithms for face recognition using FERET, ORL Yale and YaleB databases. This algorithm performs better than some of the existing popular algorithms.