一种基于运动边缘检测的视频对象分割新算法

利用人眼对运动(时间梯度)和边缘(空间梯度)都特别敏感的视觉特点，把帧间运动变化检测和图像的边缘检测结合起来，提出一种新的运动视频对象分割算法．通过帧间差快速得到运动物体的大致位置，形成差分模板，然后通过边缘检测在差分模板中确定物体的准确边缘，并形成边缘模板．在边缘模板的基础上，利用二值图像下的收缩型活动轮廓算法，可以方便地得到视频对象(VOP)的闭合轮廓曲线(以便进行编码传输和基于形状的检索)．同时，设置模板缓冲区以记忆前一时刻的分割结果，从而弥补当前帧的不完整性．该算法对目标的整体运动和局部形变都有较强的适应性，且具有自动消除显露背景以及多目标自动分割包围的能力．

A Video Object Segment Algorithm Based on Moving Edge Detection

Based on the fact that human vision is especially sensitive to both the motion and edge of image, a new video object segment algorithm is proposed, which combines the inter frame difference and image gradient An approximate location of moving object can be obtained by using the difference between two continuous frames, then the exact boundary is defined according to the gradient of image, and finally an edge template is formed, from which a closed contour of VOP can be produced by using a shrinking active contour model At the same time, a template buffer is kept to remember the result of the previous frame, so as to make up for the incompletion of the inter frame diffrence of current frame The proposed algorithm is robust to the entire motion and local deformation of object, and can remove the uncovered background automatically as well as siege and segment multiple objects