Maximum a posteriori super-resolution of compressed video with a novel multichannel image prior and a new observation model

In this paper we propose a class of SR algorithms for compressed video using the maximum a posteriori (MAP) approach. These algorithms utilize a novel multichannel image prior model which has already been presented mainly for uncompressed video, along with a new hierarchical Gaussian nonstationary version of the state-of-the-art quantization noise model. The relationship between model components and the decoded bitstream is also demonstrated. An additional novelty of this framework pertains to the transition flexibility from totally nonstationary algorithms used for compressed video to fully stationary algorithms used for raw video. Numerical simulations comparing the proposed models among themselves, verify the efficacy of the adopted multichannel nonstationary prior for different compression ratios, and the significant role of the nonstationary observation term.