Multi-task Gaussian Process Regression-based Image Super Resolution

Xinwei Jiang, Jie Yang, Lei Ma and Yiping Yang

Abstract

This paper presents a novel framework for image super resolution (SR) based on the multi-task gaussian process (MTGP) regression. The core idea is to treat each pixel prediction using gaussian process regression as one single task and cast recovering a high resolution image patch as a multi-task learning problem. In contrast to prior gaussian process regression-based SR approaches, our algorithm induces the inter-task correlation for considering image structures. We demonstrate the efficiency and effectiveness of the proposed method by applying it to the classic image dataset and experimental results show our approach is competitive with even outperforms the related and state-of-the-art methods.

Session

Poster 2

Files

PDF iconExtended Abstract (PDF, 39K)
PDF iconPaper (PDF, 1180K)

DOI

10.5244/C.29.151
https://dx.doi.org/10.5244/C.29.151

Citation

Xinwei Jiang, Jie Yang, Lei Ma and Yiping Yang. Multi-task Gaussian Process Regression-based Image Super Resolution. In Xianghua Xie, Mark W. Jones, and Gary K. L. Tam, editors, Proceedings of the British Machine Vision Conference (BMVC), pages 151.1-151.10. BMVA Press, September 2015.

Bibtex

@inproceedings{BMVC2015_151,
	title={Multi-task Gaussian Process Regression-based Image Super Resolution},
	author={Xinwei Jiang and Jie Yang and Lei Ma and Yiping Yang},
	year={2015},
	month={September},
	pages={151.1-151.10},
	articleno={151},
	numpages={10},
	booktitle={Proceedings of the British Machine Vision Conference (BMVC)},
	publisher={BMVA Press},
	editor={Xianghua Xie, Mark W. Jones, and Gary K. L. Tam},
	doi={10.5244/C.29.151},
	isbn={1-901725-53-7},
	url={https://dx.doi.org/10.5244/C.29.151}
}