摘要
Ｓｐｒｉｔｅ图像是视频对象ＶＯ（ＶｉｄｅｏＯｂｊｅｃｔ）在一段图像序列中所有可见部分的组合，是ＭＰＥＧ－４提出的重要的概念之一．它利用了图像的镶嵌技术和图形的几何变换技术，极大地提高了编码效率．本文研究了ＭＰＥＧ－４的验证模型中的动态Ｓｐｒｉｔｅ的编码算法后发现，原有算法在视频对象做快速运动时，其编码效率是不能令人满意的．通过扩大全局运动估计的搜索区域、预测当前帧的运动参数和减少迭代次数等改进后，算法在视频对象作各种形式的运动时。
A Sprite which is an image composed of the pixels belonging to a visible video object throughout a video segment is an important concept proposed by MPEG 4.In this method, the image mosaicking technology and geometric transformation technology are applied to greatly improve the encoding efficiency. By researching on the online sprite encoding algorithm in the MPEG 4 verification model,it is found that the encoding results aren’t satisfied in the case of video object fast motion. The main reason is the search region limitation in the global motion estimation. Since the simply enlarging search region is difficult to ensure the right motion estimation, the prediction of the motion of current frame with the motion information of the previous frame is adopted in the paper. With three improvements of enlarging the search region in global motion estimation predicting the current frame motion and shortening the iterative number,the algorithm can achieve high coding efficiency when video object perform various motions.The experiments show that the motion estimable range is below 32 pixels in original algorithm. However, the improvement algorithm is still able to obtain the accurately motion estimation when the video object motion exceeds 100 pixels.
出处
《计算机学报》
EI
CSCD
北大核心
1999年第3期262-268,共7页
Chinese Journal of Computers
基金
国家自然科学基金
国家八六三高技术研究发展计划
