Efficient learning of image super-resolution and compression artifact removal with semi-local Gaussian processes
- Submitting institution
-
The University of Lancaster
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 282022934
- Type
- D - Journal article
- DOI
-
10.1109/TPAMI.2015.2389797
- Title of journal
- IEEE Transactions on Pattern Analysis and Machine Intelligence
- Article number
- -
- First page
- 1792
- Volume
- 37
- Issue
- 9
- ISSN
- 0162-8828
- Open access status
- Out of scope for open access requirements
- Month of publication
- January
- Year of publication
- 2015
- URL
-
-
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
4
- Research group(s)
-
E - Interactive Systems
- Citation count
- 40
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- Learning-based approaches for image enhancement are limited in that building models is computationally intensive (requiring days or weeks of training time). This academic/industrial collaboration brings new scientific insights into the (spatially local) nature of large-scale Gaussian process, and creates an extremely efficient method that learns models on millions of images only in five minutes in a modern PC. The resulting image super-resolution and compression artifact removal algorithms have been established as standard references for recent deep neural network-based approaches. An extension of the work developed a real-time animated character control framework [Rhodin et al. SIGGRAPH 2015].
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -