An improved video object segmentation algorithm based on background reconstruction

As a critical technology for computer vision and video processing, video object segmentation has far-going pragmatism significance and application importance. In this paper, a video object segmentation algorithm based on background reconstruction is proposed to extract moving objects from video sequences, which were taken by stationary cameras. Firstly, the change detection is used to achieve the mask representing moving regions with an estimation noise parameter, then the methods of maximum in eight-neighbor regions is present to fill the interior holes. Secondly, the background image is available by mapping the mask to the correspondence frame of sequences, then the comparison of frame difference mask is adopted to rebuild the background image, in this way, video objects which have long stayed in the background will be removed from the moving regions after they turn to be stationary. Finally, the initial video object is derived in each frame by subtracting the background from this image, after that, mathematic morphology post-processing is used to get an accurate video object. Experiments on typical sequences have successfully demonstrated the validity of the proposed algorithm.