Publications‎ > ‎

Single Image Super-Resolution from Transformed Self-Exemplars



Abstract

Self-similarity based super-resolution (SR) algorithms are able to produce visually pleasing results without extensive training on external databases. Such algorithms exploit the statistical prior that patches in a natural image tend to recur within and across scales of the same image. However, the internal dictionary obtained from the given image may not always be sufficiently expressive to cover the textural appearance variations in the scene. In this paper, we extend self-similarity based SR to overcome this drawback. We expand the internal patch search space by allowing geometric variations. We do so by explicitly localizing planes in the scene and using the detected perspective geometry to guide the patch search process. We also incorporate additional affine transformations to accommodate local shape variations. We propose a compositional model to simultaneously handle both types of transformations. We extensively evaluate the performance in both urban and natural scenes. Even without using any external training databases, we achieve significantly superior results on urban scenes, while maintaining comparable performance on natural scenes as other state-of-the-art SR algorithms.

Citation

Jia-Bin HuangAbhishek Singh, and Narendra Ahuja, Single Image Super-Resolution from Transformed Self-Exemplars, IEEE Conference on Computer Vision and Pattern Recognition, 2015 (Oral)


Bibtex

@inproceedings{Huang-CVPR-2015, author = {HuangJia-Bin and Singh, Abhishek and AhujaNarendra}, title = {Single Image Super-Resolution from Transformed Self-Exemplars}, booktitle = {IEEE Conference on Computer Vision and Pattern Recognition)}, year = {2015}, volume = {}, number = {}, pages = {} }



https://uofi.box.com/shared/static/8llt4ijgc39n3t7ftllx7fpaaqi3yau0.pdf
 
Paper 
High-res [PDF] (29.7 MB)
Low-res  [PDF] (6.53 MB)
Extended Abstract [PDF]
CVPR Talk [Link]

Supplementary Material
- Comparison [PDF] (86.4 MB)
- Urban 100 [Link]     SRF 4x
- BSD 100 [Link]       SRF 3x
- Sun-Hays 80 [Link] SRF 8x
Download results
- Urban 100 [Link] (1.14 GB)
- BSD 100 [Link] (568 MB) 
- SunHays80 [Link] (311 MB)
- Set 5 [Link] (16.1 MB)
- Set 14 [Link] (86.0 MB)
- Readme [Link]




Poster 
PPT [Link] (26.5 MB)
- PDF [Link] (2.5 MB)
 https://github.com/jbhuang0604/SelfExSR


Reference code

- [GitHub page]

Urban 100 dataset - download [Link] (1.14 GB)

Sample results 4x on the Urban 100 dataset (left: bicubic, right: our result). See full comparison here.
          

Berkeley Segmentation 100 test images (BSD 100) - download [Link] (568 MB) 


Sample results 4x on the BSD 100 dataset (left: bicubic, right: our result). See full comparison here.
  
  
  
  


Sun-Hays 80 dataset - dowload [Link] (311 MB)

Sample result 8x on the Sun-Hays 80 dataset (left: bicubic, right: our result). See full comparison here.
   
 
 



Sample comparisons with the state-of-the-art super-resolution algorithms (more comparisons can be found in the supplementary material)
 

  
  



  

  
 



Comments