摘要
无论是在图象识别 ,还是在基于 MPEG- 4的图象压缩编码等应用领域 ,视频对象分割提取是其中一个很重要的技术环节 .为了在静止背景的情况下 ,能很好地解决多目标分割的问题 ,同时能进行单目标的分割 ,提出了一种鲁棒性较好的视频分割算法 .该算法通过对图象序列中每连续 3帧图象进行对称差分 ,首先检测出目标的运动范围 ;然后通过对差分结果进行聚类分析来确定该帧图象中视频对象的个数 ;接着再利用在二值差分图象上收缩的活动轮廓 ,把视频对象的轮廓精确地包围起来 ,即得到该帧分割结果 ;最后利用光流法来对视频对象进行投注跟踪、修正 .另外还利用多个图象序列对该方法进行了试验 .实验结果表明 ,在静止背景下 ,该算法无论是对运动的单目标 ,还是对运动的多目标 ,均能较好地从静止背景中分离出来 ,即能得到理想的分割结果 。
Video segmentation is an important step in image recognition and in video coding based on MPEG 4 standard. This paper proposes a robust video segmentation algorithm which can greatly segment not only multiple VOs but also single VO from the static background. From the research, it is found that the raw contours of the active objects in the sequential video frames can be gained fast and exactly by using symmetrical DFD. So first use every three continuous video frames to get symmetrical DFD, then acquire the motion area of the moving object. Second analyze the result of the symmetrical DFD by clustering to inspect the number of VOs in the video frame. During clustering analyzing, not employing the traditional fuzzy C clustering mode, but employing a new clustering mode with using the body's characteristic which is proposed firstly in this paper. Third use the active contour algorithm in the binary image to get the contours that enclose the VOs accurately. By this step acquire the result of segmentation. Last, track the VOs in the video frames by the vectors of optical flow and also modify the result in some special conditions. All tests receive the ideal segmentation effects by using this algorithm with several image sequences. The experimental results show that this algorithm can complete video segmentation both for multiple VOs and for single VO in the static background correctly, so this algorithm has some robusticity and practicality.
出处
《中国图象图形学报（A辑）》
CSCD
北大核心
2002年第11期1128-1133,共6页
Journal of Image and Graphics
基金
武汉大学软件工程国家重点实验室建设基金项目
