Moving object segmentation: A block-based moving region detection approach

This paper presents a novel block-based moving region detection algorithm to segment objects. The frame is first partitioned into homogeneous regions. Moving region is then determined by a voting procedure of pixels within the region. To exploit the local features, we divide the frame into n x n blocks and perform block analysis for moving object segmentation. An iterative motion re-estimation technique is developed to obtain reliable block motion parameters. The block eigen value is used to measure the block texture. Block location corresponding to the region partition is also considered as a clue. Based on motion, texture and location information, moving regions are classified. Experimental results show that our approach is robust and achieves remarkable performance.