Anchored Neighborhood Regression based Single Image Super-Resolution from Self-Examples

Yapeng Tian, Fei Zhou, Wenming Yang*, Xuesen Shang and Qingmin Liao

In this paper, we present a novel self-learning single image 
super-resolution (SR) method, which restores a highresolution 
(HR) image from self-examples extracted from 
the low-resolution (LR) input image itself without relying 
on extra external training images. In the proposed method, 
we directly use sampled image patches as the anchor points, 
and then learn multiple linear mapping functions based on 
anchored neighborhood regression to transform LR space 
into HR space. Moreover, we utilize the flipped and rotated 
versions of the self-examples to expand the internal patch 
space. Experimental comparison on standard benchmarks 
with state-of-the-art methods validates the effectiveness of 
the proposed approach.

Fig 1. Visual comparison of the restored HR images (close-ups) scaled by a factor of 2.

  • Source Code
    • Note that: the results might have a bit of differences with the published results due to the randomly anchor points selection scheme.
    • Baidu Cloud
    • Google Drive
    • Github


[8]    J.-B. Huang, A. Singh, and N. Ahuja, “Single image super-resolution from transformed self-exemplars,” in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, pp. 5197–5206.
[12] R. Timofte, V. De Smet, and L. Van Gool, “A+: Adjusted anchored neighborhood regression for fast superresolution,” in Asian Conference on Computer Vision (ACCV), 2014, pp. 111–126.
[13] C. Dong, C. Chen, K. He, and X. Tang, “Learning a deep convolutional network for image super-resolution,” in European Conference on Computer Vision (ECCV), 2014, pp. 184–199.