Colour Eigenfaces

摘要

Images of the same face viewed under different lighting conditions look different. It is no suprise then that face recognition systems based on image comparisons can fail when thelighting conditions vary. In this paper we address this failure by designing a new lighting conditions independent face matching technique. We begin by demonstrating that the colour image of a face viewed under any lgithing conditions is a linear transform from the image of the same face viewed under complex (3 lights at 3 locations) conditions. Our new matching technique solves for the best linear transform relating pairs of face images prior to calculating the image difference. For a database of 15 (complexly illuminated) faces and 45 test face images the new matching method delivers perfect recognition. In comparison, matching without accounting for lighting conditions fails 25