Detail-revealing Deep Video Super-resolution

TL;DR

This paper introduces a novel sub-pixel motion compensation layer within a CNN framework for video super-resolution, significantly improving detail recovery and outperforming existing methods without parameter tuning.

Contribution

The paper proposes the SPMC layer for better frame alignment in CNN-based video super-resolution, enhancing detail preservation and quality.

Findings

The SPMC layer improves alignment accuracy in video SR.

The proposed framework achieves superior visual and quantitative results.

No parameter tuning is required for high-quality outputs.

Abstract

Previous CNN-based video super-resolution approaches need to align multiple frames to the reference. In this paper, we show that proper frame alignment and motion compensation is crucial for achieving high quality results. We accordingly propose a `sub-pixel motion compensation' (SPMC) layer in a CNN framework. Analysis and experiments show the suitability of this layer in video SR. The final end-to-end, scalable CNN framework effectively incorporates the SPMC layer and fuses multiple frames to reveal image details. Our implementation can generate visually and quantitatively high-quality results, superior to current state-of-the-arts, without the need of parameter tuning.