Super-resolution applied to Distributed Video Coding with Spatial Scalability

Abstract

Distributed Video Coding is a new video coding paradigm that enables video codecs with reversed complexity. It is well-known that the performance of any distributed video coder is heavily dependent on the quality of the side information generated by motion-estimation at the decoder. As a continuation of previous works, in this paper we propose to use a super-resolution method to up-sample the non-key frames using the key frames as reference, in order to improve the side information generation in a distributed video coding scheme with spatial scalability. The framework, proposed in previous works, enables reduced encoding complexity by reduced spatial-resolution encoding of the non-reference frames, followed by Wyner-Ziv coding of the Laplacian residue. The side information generation is iterative, in the first iteration the high-frequency data of matching blocks from the key frames are added to the low-resolution blocks of the interpolated downsampled frames. Results are presented using the state-of-the-art video coding standard H.264/AVC. Keywords— Super-resolution, Wyner-Ziv, spatial scalabilty.