利用空域和时域信息自动分割MPEG-4视频对象

Automatic Segmentation of MPEG-4 Video Object Plane Using Spatial and Temporal Information

从视频序列中分割出视频对象是实现基于内容压缩编码方法的关键.该文提出了利用空域和时域信息实现MPEG 4视频对象的自动分割方法,利用相邻两帧的帧差和当前帧的纹理、颜色信息及面部定位、梯度滤波等技术,实现目标的初始化,再利用快速Hausdorff距离法跟踪目标.处理结果表明,该方法对处理相对静止的头肩序列图像具有较强的稳定性,处理速度较快,能取得较好的分割效果.

Video objects segmentation from video sequence is the key of implementing content-based compression coding method. This paper proposes a video object plane (VOP) automatic segmentation method with spatial and temporal information, which aims to initialize the objects by analyzing the difference of adjacent frames, texture information and color information of current frame, facial localization,and gradient filtering technology,and then track the objects by means of the quick Hausdorff distance method. This approach shows high stability, high speed and good segmentation when applied to relatively still head-shoulder sequences.