一种基于凸壳的压缩域运动对象快速分割算法

目前在H.264/AVC压缩域分割领域中现有的方法存在时间复杂度高,且分割运动对象不完整的问题,提出一种新的基于凸壳的压缩域运动对象快速分割(CHSTF)算法。该方法主要利用码流中的运动矢量场信息进行分割,即首先利用后向迭代累积对MV进行归一化处理,再利用时空域滤波(STF)算法对运动矢量场进行滤波得到稳定MV场,然后对滤波MV场求解凸壳并对其进行区域填充,最后对其进行优化掩膜达到分割运动对象的效果。本方法着重于快速求得整体运动对象,并获得较好分割精准度。实验表明,通过本方法能够较好地解决上述问题,并且在运动场严重缺失的环境下,相比传统方法,本方法能获得更好的效果。

A fast segmentation algorithm based on convex hull for object tracking in compressed domain

Most of the current segmentation algorithms in compressed domain are often time-consuming and have incomplete segmentation for moving objects. In this paper,based on convex hull we propose a novel fast segmentation algorithm for object tracking in compressed domain. This algorithm mainly utilizes information of the motion vector field in bits stream to do segmentation.Firstly,an iteratively backward projection scheme is proposed in normalization to obtain an accumulated motion vector filed.Then the spatio temporal filter (STF) algorithm is used to preprocess the motion vector field to obtain the stable MV field.The filled MV field is subsequently searched for building a convex hull and filling the region.Finally,the segmentation is accomplished after optimizing and forming a mask.The proposed method focuses on fast obtaining the whole moving object and a better segmentation accuracy,and the experimental results show that it can work well even when motion vector is severely lacking,and is better than the traditional methods.