| Summary: | 碩士 === 國立成功大學 === 電機工程學系 === 89 === With the development of the MPEG4, the concept of video object plane (VOP) is more and more important. Before the MPEG4 video/audio coding, it requires a prior process to decompose the video sequences into several video object planes. Each frame of a video sequence is then composed of VOP’s to the object in the scene. In this thesis, a new image sequence segmentation method for MPEG4 based on spatial-temporal information has been proposed.
The temporal segmentation obtains a coarse contour by using statistic method to separate background and non-background regions. After the temporal segmentation, the resultant mask may still be noisy due to the selection of the threshold value. We use morphological opening and closing operations to eliminate the small noises and make use of the connected-component filter to reduce the large noises. The spatial segmentation is then used to divide each image into semantic regions and find the more accurate object boundaries of moving objects. Finally, we combine temporal and spatial segmentation masks to extract the VOP’s. Experimental results indicate that the proposed method extract the objects accurately.
|
|---|
