Performance of statistical colour snakes using different colour models for multiple images

Abstract

Traditionally, the issue of colour constancy has limited the success of colour image segmentation algorithms. Statistical snakes or, more formally, active region models have been demonstrated to possess desirable characteristics when applied to image segmentation. In particular, statistical colour snakes have been shown to reduce the effects of colour constancy when a colour model, such as hue saturation value (HSV) is adopted. It is the colour constancy characteristic of the model which dictates the statistical colour snake's ability to isolate objects robustly under varying illumination conditions. The algorithm's performance therefore depend on the choice of colour space. The normalised red green blue (RGB), hue saturation value (HSV) and tint saturation luminance (TSL) colour models have all been reported as suitable candidates for achieving colour constancy. However, no comparative studies have been published on their performance within an active region model for segmenting multiple images. This paper assesses the suitability of the normalised RGB, HSV and TSL colour models as a basis for a statistical colour snake. The ability of each snake to isolate an object of uniform colour in a sequence of images under unstructured lighting is investigated and comparative experiments are performed.