基于图割的压缩域运动对象提取

Moving object extraction algorithm based on graph-cuts in compressed domain

随着在视频监控等方面的应用,视频数据量不断增加,如何快速有效地处理和分析视频内容仍然是一个亟待解决的问题。目前的运动对象提取通常采用像素域的分析方法,虽然有较好的主客观效果,但由于计算复杂度高,在实际应用中有诸多限制。因此,提出了一种基于图割的压缩域运动对象提取算法。该算法基于4×4分块的高斯背景建模,得到视频帧中各子块的初始概率,结合运动矢量(Motion Vector)信息构造压缩域图割能量函数,利用图割算法对前景区域进行修正,从而实现对运动对象的快速提取。与其他运动区域提取算法的对比实验表明,该算法具有较高的准确率和较低的计算复杂度,具有较高的实际使用价值。

With the increasing volume of video data in the applications such as surveillance, how to process and analyze video content in a fast and effective way has still been an attractive topic. The pixel-domain analysis method is widely adopted in moving object extraction. Although good performance can be achieved, there are some restrictions in the practical applications due to its high computational complexity. In this paper, a new moving object extraction algorithm based on graphcuts in the compressed domain is proposed. Background modeling is performed on the 4×4 blocks in the compressed domain so that the initial probability of each block can be obtained. Then a graph-cuts energy function can be constructed with the initial probability and Motion Vector（MV） information associated with each block. By introducing the graph-cuts algorithm to refine the foreground region, moving object segmentation can be quickly accomplished. Experimental results show that the new algorithm has high accuracy and low computational complexity which is very important for real application.