一种基于MRF的多层遗传递减视频对象分割算法

Algorithm for Video Object Segmentation Based on MRF Model Multi Layer Inherit Descending Method

提出一种视频对象分割方法,将时域信息、空域信息和图像颜色特征有效地结合起来,采用MRF遗传递减式分类来提取视频对象.利用视频帧的颜色特征,在初始分割中采用时空域分水岭方法,建立一个时空域毗连图(ST-RAG);基于时空域毗连图建立马尔可夫随机场(MRF)模型,在分割过程中采用遗传递减式方法对区域进行合理分类,利用形态学进行后处理,分割出感兴趣的视频对象.实验结果表明所获取的分割方法具有灵活性,且精度高.

A new method of video object segmentation is presented in this paper, which combines temporal information, spatial information and color feature effectively, adopting effective MRF inherit descending classification method to extract semantic video object. Firstly, making use of color feature of video frame, the spatial-temporal watershed method is used to segment the moving object in the original video sequence, and a ST-RAG （spatial temporal region adjacency graph） is built. And then a Markov random filed （MRF） model based on ST-RAG is built. The method of using multi layer inherit descending is adopted to classify region reasonably in the process of segmentation. Finally, the morpho- logical operations are used for post processing and segmenting the interesting object. Experimental results show that the selected method has the characteristics of flexibility and high precision.