Face Illumination Processing Using Wavelet Transform and Gradientfaces

Abstract

In view of the problem of illumination, a new approach based on wavelet transform and gradientfaces for illumination processing is presented. Firstly, the method calculates wavelet transform and multilevel wavelet decomposition in logarithm domain. The low frequency coefficient is discarded partially. After reconstructed, the high frequency component of the face image is enhanced using gradientfaces. Secondly, the face invariant feature is extracted by Principal Component Analysis (PCA), the nearest neighborhood classifier using cosine distance is adopted for face classification. The experiment result on Yale B frontal face database demonstrates that the presented algorithm could recognize all of the test samples and the face recognition rate is 100%. Thus, this technique can overcome the effect of different degree of illumination efficiently.