Stable Interest Point Detection under Illumination Changes Using Colour Invariants

Stable interest point detection is relevant for many computer vision applications. However, most detectors are sensitive to illumination changes, as their response varies with image contrast. In the best case, detection stability is increased using a simple image formation model assuming that illumination effects cause slowly varying changes in the image. This does not accurately model shadows and shading (interaction between illumination and scene geometry). Therefore, a new detection method is presented here, which is based on the very popular Harris detector and on the m space [5]. It yields a detection which is invariant to shadows, shading and illumination colour for matte surfaces. A preprocessing scheme is proposed to reduce the sensitivity to colour artifacts caused by demosaicing. The new detector is evaluated on real images acquired under different illuminations by comparison with other interest point detectors. The detection stability is well enhanced, especially for scenes with complex geometry.