Fast edge-directed single-image super-resolution

Mushfiqur (Nasa) Rouf1    Dikpal Reddy2    Kari Pulli2    Rabab Ward1
1University of British Columbia        2Light co


We present a novel method for single-image super-resolution (SR). In natural images, spatial edges usually have smooth contours. From this observation, we derive a fast edge-preserving natural image prior using our proposed fast edge-directed interpolation (EDI) method, and combine this prior with the well-known sparse gradient prior into a maximum-a-posteriori (MAP) formulation of the SR problem. We develop an efficient primal-dual algorithm to solve the inverse problem. The application of our edge-preserving prior adds little computational overhead and the output produced by our method demonstrates that results are better than those of the state-of-the-art methods.

Presented at

Electronic Imaging 2016, Image Processing: Algorithms and Systems XIV in San Francisco, CA, Feb 14th-18th, 2016,
as an oral presentation.


Paper: [pdf] | (external link)
Presentation slides: [ppt]

We have won the best paper award!