CS631 Paper Response Form

Paper Title: MPEG: A Video compression standard for Multimedia applications

Author(s): Didier LeGall

Main Point(s):

The paper describes the MPEG standard for encoding video and leaves issues of sound and the interleaving of sound and video aside. MPEG-video is using 3 types of frames: reference frames (called intraframes) that are basically moderately compressed image frames that serve as access points for random access of the video stream, and two types of frames that interpolate between 2 intraframes, namely 'P frames' ,which are interpolations that are encoded in reference to the preceding intraframe and bidirectionally predicted pictures (so-called B frames) that are used to interpolate images between I and P frames by encoding the predictions with reference to the preceding and the following frame. These 3 types of frames form sequences of the form IBBPBBP..BBIBB.. A motion compensation scheme is used for predicting P and B frames: for a 16x16 block, one or two motion vectors and an additive correction term are computed.Spatial redundancy is exploited similar to JPEG.

Possible Use(s):

This knowledge is useful for anybody editing or processing MPEGed video data, because one has to understand how all these predictions work if one wants to access a particular frame without decompressing the whole video.

Extensions:

One could try out how Ramin Zabih's and Dan Huttenlocher's motion algorithms influence the speed and quality/compression rate of MPEG. Also, the fact that MPEG has already computed motion (at least crudelly by assuming that the whole 16x16 block moves evenly) could be exploited by higher level processing (for example tracking of objects in movies).One could just take the MPEG computed motion vector(s) as a first approximation and do a fine-grained search exploring only small deviations from this basic motion vector.

CS631 home page