Color features show robustness against many variations such as translation, rotation, viewpoint
change, partial occlusion, low resolution, pose variations, etc. Thus, they are considered effective
cues for object representation and are widely employed for visual tracking. Mean shift algorithm is
a robust non parametric technique that is used for estimating the gradient of a density function. It is
employed widely as a fast and robust object tracker that can utilize any feature space such as the
color space. In this article, we present a simple but rather effective enhancement to the mean shift
algorithm to distinguish an object from its background by using a proper color space channel that is
selected according to the region of interest.