视频对象分割与两种面向对象的视频编码器

Video Object Segmentation and Two Kinds of Object-Oriented Video Coder

在基于对象的视频编码中 ,视频对象的分割是重要的任务 .本文研究一种利用位移帧差 (DFD)的高阶统计特性和数学形态学算子的视频对象自动分割方法 .这种方法首先根据一组转移帧差的高阶矩来得到一个大体覆盖运动对象的区域 (模板 ) ,接着用形态学的腐蚀算子从模板的边沿向内腐蚀 ,直到对象的边沿 .提出一种简单而高效的基于查找人头肩像轮廓最大转折点的头肩分离算法 .在分割的基础上 ,用软件实现了一种基于MPEG 4的视频编码系统 .提出一种面向对象分配带宽 (OOBA———object orientedbandwidthallocation)的极低比特率视频编码器 .该编码器与传统基于帧的视频编码器相比 ,在低比特率环境下 ,PSNR略为下降 ,但图像的主观视觉质量得到提高 .

Video object segmentation is an important task in object based video coding.In this paper,an automatic video segmentation scheme based on higher order statistics (HOS) of displaced frame differences (DFD) and mathematical morphological operators is studied.In this approach,a map (mark) overlapping object is first gotten by thresholding a group of higher order moment maps of DFD,and then a morphological erosion from the boundary of the mark up to the boundary of the object is performed to get the final segmentation map.A simple but efficient algorithm for separating head and shoulder is proposed based on searching for the greatest degree turning point in head shoulder contour curve.Based on segmentation,the MPEG 4 based video coding system we implemented with software is described.Then,a very low bit rate video coder with object oriented bandwidth allocation (OOBA) technology is proposed.Compared with conventional frame based video coder,the OOBA coder can improve subjective quality of images at low bit rate,while decrease PSNR just a little.