A study of the illumination cones method for face recognition under variable illumination

Abstract

Earlier face recognition algorithms have performed sufficiently well under tight environment constraints. However, simplistic and crude algorithms such as the geometric feature-based matching method and the template matching method fail catastrophically under slight environment changes, such as illumination variations. More sophisticated approaches like the Eigenface method or other Appearance-Based Methods, which had worked extremely well for face images with considerable changes in facial expression and occlusions, failed to tackle the issue of lighting changes. More recently, developments of methods that involve creating generative models of face images and extracting three-dimensional shape of faces have shown promising results. This report shall attempt to provide a detailed account of the Illumination Cones method that has shown a vast accuracy improvement compared to earlier methods in terms of robustness against illumination variation