一种半自动分割视频对象的方法

提出一种半自动视频对象分割方法，通过对跟踪分割视频序列的后继帧，这种方法首先采用基于块匹配和最大边缘强度的运动估值和补偿方法进行对象轮廓定位，接着采用模板匹配以特定对象知识检测对象像素，为使轮廓定位更可靠，在块匹配的运动估值中使用了彩色信息，而模板匹配则使分割结果精确化，避免误差传递，并且在出现遮挡时只要对象颜色在整个序列中一直保持相似性，就能够正确测出对象，实验结果证明这种方法能够分割复杂场景中的任意对象。

A Semiautomatic Video Object Segmentation Method

A new semiautomatic video object segmentation method is presented. In the object tracking process to segment the subsequent frames of a video sequence, object contours are localized by using motion estimation and compensation based on block matching and maximal edge strength, and then object pixels are detected by using template matching with object-specific knowledge. To make the contour localization more reliable, color information is further used in the block matching motion estimation. The template matching refines the segmentation results, avoids error propagation and detects objects correctly in the presence of occlusion as long as object colors maintain similarity over a video sequence. Experimental results show this approach is able to segment non-rigid objects from complex scenes.