COLOR FACE IMAGE ENHANCEMENT BASED ON SVD OF DCT COEFFICIENTS IN LOGARITHM DOMAIN

Abstract

In this paper, we provided an effective illumination compensation method to
improve the quality of human face images in RGB color space for face recognition.
First, three color channels of color face image and Gaussian reference are converted
to the logarithm domain by logarithm transform (LT), then using the discrete cosine
transform (DCT) to determine their low-frequency components. The compensated
factors corresponding to the three RGB color channels will be calculated
automatically through the ratio of the largest singular values of the low-frequency
DCT coefficient matrices of the three color channels and the Gaussian reference.
Experimental results on two databases, namely CMU-PIE and Color FERET show that
our method can improve the performance of the face recognition system, better
than relevant studies such as ASVD and TSVD.