An Efficient Similarity Metric for Pattern-Based Very Low Bit-Rate Video Coding

Abstract

In the context of very low bit-rate video coding, pattern representation of a moving region (MR) in block- based motion estimation and compensation has become increasingly attractive for its very high compression and real time applications. Generally, all pattern-matching algorithms apply a similarity metric, involving elementary operations, to compute the pixel-wise difference between an MR and a particular fixed pattern in order to select the best-matching pattern for the MR from a fixed-size codebook of predefined patterns. The existing Pattern Included Similarity Metric (PISM) considers the non- overlapping areas of both the MR and pattern to compute the overall mismatch. This paper theoretically establishes that considering only the non-overlapping area of the MR is sufficient to compute the same similarity measure. Based on this, a new Pattern Excluded Similarity Metric (PESM) is developed, which is not only faster but also efficient in coding as it improves the intermediate filtering process of identifying MRs suitable for pattern representation. Empirical results show that while the PESM was used (instead of the PISM) in the latest pattern based coding, the computational coding efficiency was improved by as much as 28%, leading to a speed up as high as 74% compared with the H.264 digital coding standard while maintaining a better PSNR for the entire very low bit-rate range.