Improved illumination-invariant descriptors for robust colour object recognition

Colour object recognition is heavily influenced by variation in the scene illumination conditions. This paper proposes a set of illumination-invariant descriptors of image content. The descriptors are based on a moment-based approach to histogram comparison and, in the case of an object imaged under two different lighting conditions, permit straightforward recovery of the illumination change involved. The efficacy of the descriptors is compared experimentally with a variety of existing techniques, using an established methodology and an existing purpose-built dataset. The evidence suggests that the new descriptors outperform existing techniques in the area of colour object recognition.