Complementing SRCNN by Transformed Self-Exemplars

Super-resolution algorithms are used to improve the quality and resolution of low-resolution images. These algorithms can be divided into two classes of hallucination- and reconstruction-based ones. The improvement factors of these algorithms are limited, however, previous research [, ] has shown that combining super-resolution algorithms from these two different groups can push the improvement factor further. We have shown in this paper that combining super-resolution algorithms of the same class can also push the improvement factor up. For this purpose, we have combined two hallucination based algorithms, namely the one found in Single Image Super-Resolution from Transformed Self-Exemplars and the Super-Resolution Convolutional Neural Network from . The combination of these two, through an alpha-blending, has resulted in a system that outperforms state-of-the-art super-resolution algorithms on …

Recommended citation: Aakerberg, Andreas and Rasmussen, Christoffer B and Nasrollahi, Kamal and Moeslund, Thomas B (2016). "Complementing SRCNN by Transformed Self-Exemplars." International Workshop on Video Analytics for Audience Measurement in Retail and Digital Signage.
Download Paper