Moving cast shadow detection using joint color and texture features based on direction and distance

Moving cast shadow detection, which is a crucial and challenging problem in the community of image processing, has long been widespread concern. To solve the problem, this paper proposes a novel multi-stage approach. The proposed approach includes two main steps. The first is to select the candidate shadow pixels or regions, and the other is to achieve an accurate detection by using joint color and texture features based on direction and distance comprehensively. Both qualitative comparisons and quantitative evaluation illustrate that our approach could obtain finer performance, which achieves 86.4% of average F-measure among six typical scenes, removing moving cast shadow effectively and preserving foreground object well, especially for outdoor scenes.