A Color Similarity Measure for Robust Shadow Removal in Real Time

We introduce an approach for realtime segmentation of a scene into foreground objects, background, and object shadows superimposed on the background. To segment foreground objects, we use an adaptive thresholding method, which is able to deal with rapid changes of the overall brightness. The segmented image usually includes shadows cast by the objects onto the background. Our approach is able to robustly remove the shadow from the background while preserving the silhouette of the foreground object. We discuss a similarity measure for comparing color pixels, which improves the quality of shadow removal significantly. As the image segmentation is part of a real-time interaction environment, real-time processing is needed. Our implementation allows foreground segmentation and robust shadow removal with 15 Hz.