Clustering-based shadow edge detection in a single color image
Wang Shiting, Hong Yuan Zheng · 2013
Shadow edge detection is an important and challenging part of shadow detection and removal. In this paper, we propose a clustering-based shadow edge detection method, which can avoid choosing parameters by the hysteresis step in the Canny edge detection process in Finlayson's method. First, K-means clustering is applied to the derivative difference of the brightness image and light-invariant image from raw input. Second, punishment rules are exploited to correct false alarms. Plus, putting off morphological dilation prevents from introducing extra material edges into shadow edge mask. Experimental results show that compared with Finlayson's work, our method can effectively generate complete enough shadow edge mask. Importantly, parameters of our method are not affected by Canny detection and Mean-shift smoothing processing, thus producing robust and stable results in both indoor and outdoor scenes.