Motion Compensated Object Based Video Coding

本研究计划研究一种用于多媒体、网络视频、视频会议和视频点播等应用的高级视频压缩方法。它是通常用于极低速率的基于对象的编码和通常用于较高速率的基于3-D子带/小波波形的编码的新组合。在该方法中，首先将运动场分割成组成图像帧的对象，并以不同的速度运动。使用保持边缘形状的轮廓编码类型对这些对象的边界进行编码。该轮廓编码作为针对前一帧找到的对象边界的更新来执行。最后利用基于运动补偿三维子带滤波的波形编码方法对这些空时目标的内部进行编码。总体目标是一套高效、统一的视频压缩算法，可从视频电话扩展到高清电视。

This research program investigates an advanced video compression approach for applications such as multimedia, video on networks, video conferencing, and video on demand. It is a novel combination of object-based coding usually used for very low rates and 3-D subband/wavelet waveform-based coding, usually reserved for higher rates. In this method, the motion field is first segmented into objects which make up the image frame, and move at different velocities. The boundaries of these objects are coded using a type of contour coding which preserves the shape of the edge. This contour coding is performed as an update on the object boundaries found for the previous frame. Finally the interior of these space-time objects is coded using the waveform coding method based on motion compensated 3-D subband filtering. The overall goal is an efficient unified set of video compression algorithms scalable from video phone up to HDTV.