Super Resolution using Edge Prior and Single Image Detail Synthesis

Yu-Wing Tai

Shuaicheng Liu

Michael S. Brown

Stephen Lin

Abstract¡ªEdge-directed image super resolution (SR) focuses on ways to remove edge artifacts in upsampled images. Under large magnification, however, textured regions become blurred and appear homogenous, resulting in a super-resolution image that looks unnatural. Alternatively, learning-based SR approaches use a large database of exemplar images for ¡°hallucinating¡± detail. The quality of the upsampled image, especially about edges, is dependent on the suitability of the training images. This paper aims to combine the benefits of edge-directed SR with those of learning-based SR. In particular, we propose an approach to extend edge-directed super-resolution to include detail from an image/texture example provided by the user (e.g., from the Internet). A significant benefit of our approach is that only a single exemplar image is required to supply the missing detail ¨C strong edges are obtained in the SR image even if they are not present in the example image due to the combination of the edge-directed approach. In addition, we can achieve quality results at very large magnification, which is often problematic for both edge-directed and learning-based approaches.


Source Codes (Partial) in C++

Related Project:
Colorization for Single Image Super Resolution
Perceptually-Inspired and Edge-Directed Color Image Super-Resolution


author = {Yu-Wing Tai and Shuaicheng Liu and Michael S. Brown and Stephen Lin},
title = {Super Resolution using Edge Prior and Singale Image Detail Synthesis},
booktitle = {CVPR},
year = {2010}

Nearest Neighbor
LR-RMS 0.60 HR-RMS 11.87
LR-RMS 0.61 HR-RMS 9.06
Back Projection
LR-RMS 3.05 HR-RMS 10.66
Gradient Profile Prior [CVPR'08]
LR-RMS 1.89 HR-RMS 7.64
Learning [IJCV'00]
LR-RMS 3.14 HR-RMS 16.59
Our result with sand texture
LR-RMS 3.45 HR-RMS 15.89
Our result with zebra texture
LR-RMS 3.10 HR-RMS 14.85
Our result with circle image
LR-RMS 2.17 HR-RMS 7.32
Ground Truth
LR-RMS 0.00 HR-RMS 0.00
10x super-resolution on a synthetic example. Our approach generates different results depending on the supplied texture. The lower left corner shows the result image after 10x downsampling. Note that for all results, the down-sampled images are approximately identical. Listed below each result are the LR-RMS errors (RMS errors with respect to the low resolution input), and the HR-RMS errors (RMS errors with respect to the high resolution ground truth image).


Input and example image
Learning [IJCV'00]
HR-RMS 24.3 MSSIM 0.62
Alpha Channel [CVPR'07]
HR-RMS 9.3 MSSIM 0.70
Gradient Profile Prior [CVPR'08]
HR-RMS 8.4 MSSIM 0.75
Our Result
HR-RMS 10.6 MSSIM 0.77
Ground Truth

Our 10x magnification
Face with freckles. (a-e) 4x magnification result of various approaches. (f) Ground truth. (g) Our result with a 10x magnification. The HR-RMS errors and the MSSIM score with respect to the 4x ground truth image are listed below each result.


Input and example image
Gradient Profile [CVPR'08]
Learning [IJCV'00]
Our Results