A fast method to improve the stability of interest point detection under illumination changes

This paper presents a method to improve the stability of interest point detection under illumination changes. The response of state-of-the-art detectors depends on image contrast, which leads to unstable detection, especially when the position or orientation of the light source changes. Based on a simple image formation model and on ideas of homomorphic filtering, the Harris corner detector-one of the most widely used detectors-is modified to reduce the illumination influence. The computation effort needed for detection is only minimally increased. The algorithm performances are evaluated on real images using repeatability and false positive rates. The detection stability is enhanced, above all under complex lighting changes.