A DCT-Based Video Object Segmentation Algorithm for Rainy Situation Using Change Detection

This paper proposes a DCT-based moving object extraction algorithm which is focused on rainy condition by change detection method. The DCT technique is used to decrease the rainy effect in order to construct a reliable background model. Then, we use change detection to classify pixels in video frame into the foreground region or background region so as to acquire an initial object mask. Besides, reflection effect can be removed from, initial object mask through edge detection and bounding box match. Finally, a post-processing is used to refine the boundary of moving object by connect component labeling and morphological operation. Experimental results show that the proposed algorithm can provide a satisfactory segmentation result in rainy situations