Adaptive coding of moving objects for very low bit rates

摘要

This paper describes an object-based video coding system with newnideas in both the motion analysis and source encoding procedures. Thenmoving objects in a video are extracted by means of a joint motionnestimation and segmentation algorithm based on the Markov random fieldn(MRF) model. The two important features of the presented technique arenthe temporal linking of the objects, and the guidance of the motionnsegmentation with spatial color information. This facilitates severalnaspects of an object-based coder. First, a new temporal updating schemengreatly reduces the bit rate to code the object boundaries withoutnresorting to crude lossy approximations. Next, the uncovered regions cannbe extracted and encoded in an efficient manner by observing theirnrevealed contents. The objects are classified adaptively as P objects ornI objects and encoded accordingly. Subband/wavelet coding is applied innencoding the object interiors. Simulations at very low bit rates yieldedncomparable performance in terms of reconstructed PSNR to the H.263ncoder. The object-based coder produced visually more pleasing video withnless blurriness and devoid of block artifacts, thus confirming thenadvantages of object-based coding at very low bit-rates