Color Angle Invariants for Object Recognition
Graham David Finlayson, S. S. Chatterjee, Brian Funt · Color and Imaging Conference · 1995
We present a fast, color-based algorithm for recognizing objects viewed under an unknown illuminant. Objects are indexed by just three numbers: the angles of the object's color distribution. If R, G and B denote the 3 color bands of the image of an object (stretched out as vectors) then the angular index comprises the 3 interband angles (one per pair of vectors). In the general case the distribution of colors, and in turn the angular index, will depend on the color of the illuminant. If, however, the original color bands are transformed by a sharpening transform before computing the distribution angles, then we show that the angular index is illuminant independent. Indexing using angles calculated post sharpening delivers excellent recognition for a variety of illuminations.