Removing shadows

Shadows often confound algorithms designed to solve computer vision tasks such as image segmentation, object detection and tracking. In this paper, a shadow detection and removal technique is proposed. The method requires no camera calibration or other a priori information regarding the scene. Support vector machines are used to identify shadow boundaries based on their boundary properties. This boundary information is used to identify shadowed regions in the image and then assign them the color of non-shadow neighbors of the same material.