In-band motion compensated temporal filtering

Abstract

A novel framework for fully scalable video coding that performs open-loop motion-compensated temporal filtering (MCTF) in the wavelet domain (in-band) is presented in this paper. Unlike the conventional spatial-domain MCTF (SDMCTF) schemes, which apply MCTF on the original image data and then encode the residuals using the critically sampled discrete wavelet transform (DWT), the proposed framework applies the in-band MCTF (IBMCTF) after the DWT is performed in the spatial dimensions. To overcome the inefficiency of MCTF in the critically-sampled DWT, a complete-to-overcomplete DWT (CODWT) is performed. Recent theoretical findings on the CODWT are reviewed from the application perspective of fully-scalable IBMCTF, and constraints on the transform calculation that allow for fast and seamless resolution-scalable coding are established. Furthermore, inspired by recent work on advanced prediction techniques, an algorithm for optimized multihypothesis temporal filtering is proposed in this paper. The application of the proposed algorithm in MCTF-based video coding is demonstrated, and similar improvements as for the multihypothesis prediction algorithms employed in closed-loop video coding are experimentally observed. Experimental instantiations of the proposed IBMCTF and SDMCTF coders with multihypothesis prediction produce single embedded bitstreams, from which subsets are extracted to be compared against the current state-of-the-art in video coding.