Decision trees for error concealment in video decoding

Abstract

When macroblocks are lost in a video decoder such as MPEG-2, the decoder can try to conceal the error by estimating or interpolating the missing area. Many different methods for this type of post-processing concealment have been proposed, operating in the spatial, frequency, or temporal domains, or some hybrid combination of them. In this paper, we show how the use of a decision tree that can adaptively choose among several different error concealment methods can outperform each single method. We also propose two promising new methods for temporal error concealment.

Error Concealment for Dual Frame Video Coding with Uneven Quality

“…One significant conclusion is that the decision tree approach gives a performance advantage over not using it, even when the tree is not customized for the particular sequence, but rather is constructed from unrelated sequences. This is in contrast to the work reported in [8] where a CART decision tree (for a single frame codec) was used to choose among spatial, temporal, and frequency concealment, and it was found that performance gains could be had only by a customized tree. Because it required a customized tree, that system required side information and could not be standard compliant.…”

“…A further level of adaptivity appears in [6] and [7]. In [8], a CART decision tree is used to decide upon a particular EC approach among several EC algorithms.…”

Spatiotemporal error concealment with optimized mode selection and application to H.264

“…The importance of mode selection on the quality of the resulting video should not be underestimated [7], since spatial and temporal concealment provide quite different visual quality and PSNR. If the MVs can be estimated with a satisfactory degree of accuracy, temporal concealment usually yields better results, since the replacement MB is sharp and fairly similar to the original.…”

Comparison of error concealment strategies for MPEG video

“…Of the 8 methods listed in Table 1, none is consistently best for all MBs. In our previous work [2], the EC method is chosen by a decision tree which looks at the context of the missing MB. The tree is designed at the encoder using the CARTTM algorithm [ 11, and is transmitted to the decoder.…”