Shading correction and segmentation of color images

A simple nonparametric analysis of feature space has been employed for segmentation of shaded color images. The proposed segmentation method in the 3D RGB color space is based on the well-known mean shift algorithm, which was modified to a max shift algorithm, and cluster growing procedure. The method requires a simple initialization, ie, designation of one object point in the image. The max shift algorithm then examines the cluster appending suitability of feature space points in a cluster growing procedure. In this way, problems of parametric methods with complicated cluster shapes are avoided. Color shading is suppressed by an information-theoretic method by which the information of each color component is separately minimised. The generality of the proposed method is shown by various examples. Moreover, the method proves to be robust and fast and as such may be applied for various tasks in visual inspection of colorful scenes.