A novel shadow elimination algorithm for moving object segmentation

Elimination of cast shadow is one of the most challenging problems in the computer vision. In this paper, we present a novel approach to segment moving objects by shadow flow with GMM and graph cuts. We model the cast shadow on a pixel as shadow flow models that can describes moving cast shadows on surfaces, and model intensity of each pixel as GMM. General algorithm classifies each current pixel to shadow flow, background and foreground. For Moving Object Segmentation, foreground object is the most interest, so we do not tell shadow pixel and background pixel, just tell foreground pixel and no-foreground pixel. A graph containing image information is constructed, in which different weight are assigned to graph edges according to difference between current pixel and background models, shadow flow models. Finally, a dynamic graph cuts algorithm is used to segment the moving objects from background. Experiments show that our approach achieves excellent performance.