摘要： 针对固定场景的视频，提出一种基于时空分割的视频分割算法。该算法在时域中利用Tophat形态学滤波得到视频对象的精确位置，在空域中采用基于t混合模型和贪婪EM的聚类算法进行单帧图像分割。将时域定位和空域分割结果结合，可以准确地将视频序列中感兴趣的运动目标分割出来。实验表明，该算法能够得到完整的视频对象，有一定的理论意义和实用性。

关键词: 时空联合, Tophat形态学滤波, 有限混合模型, 贪婪EM

Abstract: This paper proposes an algorithm of video segmentation based on spatio-temporal information for the videos with fixed scenes. In the time domain, it makes use of morphological Tophat filtering to obtain the accurate position of video objects, and in the spatial domain, a clustering algorithm based on t-mixture model and Greedy EM is used to perform segmentation for single frame. By combining the time domain location with the result of spatial domain segmentation, it can correctly obtain the interesting moving target from the video sequence. Experiments show that this algorithm can get better results for theory and application.

Key words: spatio-temporal combination, morphological Tophat filtering, finite mixture model, Greedy EM

中图分类号: 

• TP301.6