Study of moving-object segmentation in DCT-based compressed video

Abstract

This dissertation mainly concerns moving-object segmentation in DCT domain. Video segmentation, or moving-object segmentation, is essential for the content-based functionalities of recent digital signal processing applications. Since the current video coding standards adopt block-based compression, the proposed segmentation algorithm is block-based, and it uses the motion vector and DCT coefficients of the block for segmentation features. To overcome the demerit of block-based operation, the proposed segmentation algorithm introduces statistical hypothesis testing. The proposed algorithm is comprised of three techniques, such as moving-object segmentation in compressed bitstream, moving-object segmentation with static sprite in the video source coder, and moving-object segmentation with adaptive sprite. In the compressed bitstream, moving-objects can be segmented without decompression. In the video source coder, moving-objects are segmented in parallel with coding process without any previous segmentation information, where the proposed segmentation algorithm uses static sprite or adaptive sprite as a segmentation assistant. With the sprite coding, there are improvements in the coding efficiency as well as the segmentation accuracy. Though the proposed algorithm segments moving-object with block-resolution, it finds all blocks including pixel-resolution moving-object. Moreover, the proposed algorithm can reduce bitrate up to half of the conventional coding without the proposed segmentation depending on object. In addition, it’s video quality is the same as that of conventional coding. Since the proposed algorithm is an automatic segmentation, it is more useful in real-time applications than semi-automatic segmentation. The moving-object segmentation in compressed bitstream can be applied to video transcoding in communication networks and broadcasting channels. The moving-object segmentation in the video coder can be applied to surveillance and video conferenci...