Content-based Video Object Segmentation Technique

The content-based video object (VO) segmentation is the research focus of the digital video processing technology and the computer vision. Although MPEG-4 defines VO, the concrete method of VO segmentation hasn't been explained, which becomes the topic investigated in detail in this thesis. The main work includes:. The actuality of video object segmentation is studied. The advantages, shortcomings and the correct field for each one of the representative methods nowadays is also analyzed and discussed.. The basal technologies needed in video object segmentation have been investigated deeply.The highlights and the new method proposed:1 . An automatic video object generation method with boundary block segmentation has been proposed. For the video object segmentation in the head-shoulder sequences with a quiescent background, the experiment shows that the method proposed gets a good result in subjective visual effect. It avoids some general problems with the traditional segmentation methods, such as the over-segmentation with mathematical Morphology, the instability and aperture effect with optical flow.2. The multi-scale layer idea of the pyramid theory is successfully imported in this thesis, so a segmentation pattern comes into being, which follows upper layer to lower layer and blocks to pels. Tested with the universal head-shoulder video—Claire sequences, this pattern can process 12 frames per second (F/S). Compared with the rates of the other international representative methods, such as 4F/S with Kim's method and 7F/S with Li's method, the pattern proposed accelerates remarkably.3. The traditional threshoding method is improved to get the right threshold for the image binarization. The wrong blocks and the missing blocks are limitedto one block.