Performance analysis of DCT in logarithm domain and two -point normalization method for illumination and expression variation in face recognition

Illumination and expression variation are the major challenges in the face recognition. This paper presents comparative analysis of two normalization techniques namely, DCT in Log domain and 2-point normalization method.. The DCT is employed to compensate for illumination variations in the logarithm domain. Since illumination variation lies mainly in the low frequency band, an appropriate number of DCT coefficients are truncated to reduce the variations under different lighting conditions. The nearest neighbor classifier based on Euclidean distance is employed for classification. The 2-point normalization method considers only center points of eyes. Preprocessing is done using the Gaussian filter. Based on the center points of eyes the rotation of image is performed. Rotated image is masked using an ellipse. Histogram equalization is performed on the unmasked part of the image. The Log-Gabor filter of 5 × 5 window size is used to extract the features. The Cosine based distance method is used for classification. Experimental results on the Yale B and Cafe database show that the proposed approach of normalized dataset improves the performance significantly for the face images with large illumination and expression variations. The face recognition system based on Log-Gabor filter achieves the recognition accuracy of 85% to 93%, and that based on DCT normalization achieves the recognition accuracy of 97% to 100% using the above databases.