Compressing Surveillance Video using Motion Segmentation

In this paper, we propose a method of compressing surveillance video that, given a foreground segmentation, encodes the regions of interest using a block-based approach that reduces the distortion due to segmentation errors. Compression of each region of interest, as well as the background image, is achieved using JPEG 2000. With an estimate of the background image being updated periodically, each image is reconstructed by combining the regions of interest with the most recently encoded background image. Experiments demonstrate that using our method can result in greater compression and less distortion than other frame based compression methods.