An Improved Cast Shadow Detection Method with Edge Refinement

This paper presents an improved shadow discriminating method by dividing the cast shadow region into sub regions. We analyze the difference caused by distinct illumination condition for each sub shadow region, and then propose a classification method. We divide cast shadow into deep umbra shadow, shallow umbra shadow and penumbra shadow through gray-level and color features. For color similarity examination, we raise a new thresholding algorithm on color shift angle and give the simulation result. Compared to classic method considering cast shadow as a whole region, our new method focuses on locating sub shadow regions separately and improves accuracy. After that, we present a refinement algorithm to find fake shadow regions in previous detection result. In this step, we utilize a unique feature of cast shadow on edge image to separate the vehicle edges and shadow edges