视频对象分割在基于内容的视频编码和视频检索中均有重要的应用.为此,针对视频对象分割,提出了一种时域和空域信息融合的视频对象分割方案,该方案首先对时域分割采用基于F-假设检验的方法来得到初始的变化检测模板,然后通过与基于形态学的空域分割融合来获得最终的运动对象.实验结果表明,该方案计算比较简单,能较好地将前景运动对象从静止或运动、简单或复杂的背景中分离出来,且定位精度较好.

Video object segmentation has important application in content-based video encoding and video retrieval. In this paper, a new approach based on temporal and spatial union information is proposed. In spatial domain, morphological operators such as morphological filters, morphological gradient operator and watershed algorithm are used, morphological filters are used to simplify the image, morphological gradient is used to mark homogeneous regions, the watershed operator is used to decide spatial boundaries. In temporal segmentation domain, the initial change detection mask is gotten by using F-test, then by combining it with the result of spatial segmentation, the final result is gotten. The procedure toward complete segmentation consists of the following main steps; global motion estimation and compensation (if exist background motion), change detection, marker extraction, watershed transfer, temporal and spatio information fusion. The experimental result shows the effectiveness of the method: it can separate the foreground from still or motion, simple or complex background satisfactorily, preferably locate and segment the video object.