一种视频对象分割技术的研究与实现

本文提出一种可以从复杂场景中分割出视频对象的方法.首先对视频图像进行全局运动估计和补偿校正,并将相邻两帧进行差分检测.然后利用Neyman-Pearson准则从差分图像中提取运动区域,根据提取的运动区域研究了一种在序列图像差分域中进行区域生长的算法,获得了视频对象的二值化分割模板.最后利用运动检测算法剔除由于噪声导致的标记块.实验结果表明,本文所述方法能够有效的分割出视频对象.

Research and Implementation of A Video Object Segmentation Technology

In this paper a technology of video object segmentation is presented. Firstly, the current frame is compensated after globe moving estimation, and the difference operation is made between the current frame that has been compensated and the prior frame; then the moving region is obtained from the difference image based on Neyman-Pearson criteria, and in the difference image a new region growing algorithm which overcome the default of the traditional algorithm of region growing in the time-domain is used to obtain the binary segmentation model of video object; finally, based on the moving detection algorithm the labeled blocks that are resulted from noise are eliminated. The experiment result shows that the technology can segment the video object efficiently.