一种有效的基于时空信息的视频运动对象分割算法

摘要

为实现视频编码标准MPEG-4中语义对象的自动提取,提出一种基于时空信息的运动对象分割算法.在时域上通过双边加权累积帧差和分块高阶统计算法得到目标的运动区域检测模板,以在充分利用时域信息的同时提高算法的速度；在提取空域信息时,先对视频序列的灰度图进行对比度增强处理,然后利用自适应Canny算子获取准确的空间边缘信息；最后进行时空融合,用空域边缘信息修正过的时域运动模板来提取运动对象.实验结果表明,本算法可以快速准确地分割视频运动对象.%To achieve the automatic extraction of semantic objects in video coding standard MPEG-4, this paper proposed a moving object segmentation algorithm based on temporal-spatial information. The proposed algorithm firstly obtained the moving region detection template of the target in temporal domain by using the bilateral weighted cumulative frame difference and blocked higher-order statistics algorithm, and it increased the speed of the algorithm while taking full information of temporal. Secondly, in the extraction of spatial information, it enhanced the contrast of the grayscale in video sequence at first, and then obtained accurate information of spatial edge by using an adaptive Canny operator. Finally, the proposed algorithm merged the temporal information and spatial information together, extracted the moving objects with the temporal moving template which had been corrected by the spatial edge information. The experimental results show that the proposed algorithm can obtain an accurate segmentation of moving objects in video with a high speed.