基于模板匹配的视频对象分割  

Video Object Segmentation Based on Template Matching

作　　者：宋立锋[1] 韦岗[1] 王群生[1] 

机构地区：[1]华南理工大学电子与信息学院,广东广州510641

出　　处：《电子学报》2002年第7期1075-1078,共4页Acta Electronica Sinica

基　　金：国家自然科学基金;教育部回国人员科研基金

摘　　要：视频对象分割是MPEG 4标准关键技术 .本文结合模板匹配和基于运动估值和补偿的对象跟踪方法 ,提出了一种可以从复杂场景中分割出MPEG 4视频对象的新方法 .在使用运动估值和补偿得到分割掩膜后 ,以初始帧对象颜色为模板 ,在当前帧的轮廓边界区域通过模板匹配检测对象 ,使轮廓精确化 .本文方法在一定范围内有效解决了遮挡问题 ,并能够以初始帧跟踪任意长序列中的对象 .Video object segmentation is the key issue in the MPEG-4 International Standard.This paper presents a new method for MPEG-4 video object segmentation from complicated scenes.In this method, the method of template matching and the object tracking method based on the motion estimation and compensation are combined.After the segmentation mask of the current frame has been generated by the motion estimation and compensation,the object colors in the initial frame are the same as the template and the object is detected in the object boundary area by template matching.So the object contour is refined.This object tracking method is able to solve the problems of occlusion and complicated scenes efficiently to certain extent and also tracks objects from an initial frame over any long time.

关 键 词：视频对象分割 遮挡 初始帧 对象跟踪 模板匹配 MPEG-4 

分 类 号：TN919.81[电子电信—通信与信息系统]