Shadow Invariant Segmentation based on Mean Shift

In this paper, a nonparametric shadow invariant segmentation technique is addressed for the analysis of an illuminated image as well as the segmentation of arbitarily shaped cluster. A change of illumination is a traditional and serious problem in the field of image processing. We aim at a solution to this illumination changing problem. The basic computational module of the proposed algorithm is a mean shift. Mean shift, a simple iterative procedure that shifts each data points to the average point in its neighborhood, is a class of mode detection and clustering algorithm using the gradient of a probability density function. This mean shift algorithm is applied to global HSI histograms in order to determine the natural color of objects. The proposed algorithm is invariant to the changes of illumination, e.g. shadow and shade. To show the validity of the proposed algorithm, some examples are given.