Achieving brightness-insensitive measurements of colour saturation for use in object recognition

Object recognition using colour images has great potential due to the greater information content of colour images compared with monochrome and the falling cost of colour hardware. One particular advantage of colour processing in object recognition is the potential to eliminate segmentation problems caused by shadow, on the assumption that the colour properties of an object are independent of the illumination intensity. However, there are difficulties with colour which the authors have been researching. In particular, the elimination of colour variation due to shadows requires correct colorimetric camera calibration and a colour pixel representation which separates the colour properties from the brightness. The traditional hue and saturation representation has been studied, and although it is possible for hue to be made brightness invariant by correct camera calibration, an equivalent brightness invariance of saturation is much more difficult to achieve. In this paper we present an alternative measure of depth of colour which we call colour purity. We define colour purity, P_c , as follows: P_c=r+g+b- 3.min(r,g,b). This quantity is easily computed and exhibits greater brightness invariance than traditional saturation in some circumstances. Use of colour purity can facilitate segmentation of colour images using colour properties and cluster analysis. Specifically, its potential to limit the effects of quantisation noise in colour image processing is described