Illumination normalization using down-scaling of low-frequency DCT coefficients in DWT domain for face recognition

In this paper, an approach for illumination normalization based on DWT and DCT for face recognition under varying lighting conditions is presented. The wavelet transform decomposes the full image into one low-frequency and three high frequency sub bands. To compensate the illumination variations, it is required to suppress the low-frequency components. At the same time, the useful information from the low-frequency components has to be preserved for the purpose of discrimination between different face images. Therefore, in the presented approach, firstly, DWT is performed on the image and DCT Transform is employed on low-frequency sub band. Then Low-Frequency DCT (LFDCT) coefficients are modified to suppress the illumination variations. Finally, improved low-frequency band and rest three high frequency sub bands are combined using inverse DWT. Classification is performed using k Nearest Neighbor Classifier (k-NNC). To show the performance improvement of the given approach, experiments have been performed on Yale B database. Significant reduction in error rate has been achieved by the given approach of down scaling low-frequency DCT coefficients of low-frequency sub band of DWT.