Fast and Accurate Image Super-Resolution Using a Combined Loss

Jinchang Xu, Yu Zhao, Yuan Dong, Hongliang Bai; The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2017, pp. 97-103

Abstract


Recently, several methods for single image super-resolution(SISR) based on deep neural networks have obtained high performance with regard to reconstruction accuracy and computational performance. This paper details the methodology and results of the New Trends in Image Restoration and Enhancement (NTIRE) challenge. The task of this challenge is to restore rich details (high frequencies) in a high resolution image for a single low resolution input image based on a set of prior examples with low and corresponding high resolution images. The challenge has two tracks. We present a super-resolution (SR) method, which uses three losses assigned with different weights to be regarded as optimization target. Meanwhile, the residual blocks are also used for obtaining significant improvement in the evaluation. The final model consists of 9 weight layers with four residual blocks and reconstructs the low resolution image with three color channels simultaneously, which shows better performance on these two tracks and benchmark datasets.

Related Material


[pdf]
[bibtex]
@InProceedings{Xu_2017_CVPR_Workshops,
author = {Xu, Jinchang and Zhao, Yu and Dong, Yuan and Bai, Hongliang},
title = {Fast and Accurate Image Super-Resolution Using a Combined Loss},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {July},
year = {2017}
}