Abstract | ||
---|---|---|
Single image super resolution task is aimed to recover a high resolution image with pleasing visual quality from a single low resolution image. It is a highly under-constrained problem because of the ambiguous mapping between low/high resolution patch domain. In order to alleviate the ambiguity problem, we split input patches into numerous subclasses and collect exemplars according to the sparse dictionary atoms. However, we observe that there still exist some similar regressors do not share the same regression in the same subclass, which may increase the super-resolving error for training data in each cluster. In this paper, we propose a robust and effective method based capped l(2,1)-norm regression to address this problem. The proposed method can automatically exclude outliers in each cluster during the training phase and give the potential to learn local prior information accurately. Numerous experimental results demonstrate that the proposed algorithm achieves better reconstruction performance against other state-of-the-art methods. |
Year | DOI | Venue |
---|---|---|
2018 | 10.1007/978-3-319-97310-4_2 | PRICAI 2018: TRENDS IN ARTIFICIAL INTELLIGENCE, PT II |
Keywords | Field | DocType |
Capped l(2,1)-norm regression, Local linear regression, Single image super resolution | Training set,Pattern recognition,Regression,Effective method,Computer science,Outlier,Local regression,Artificial intelligence,Superresolution,Ambiguity | Conference |
Volume | ISSN | Citations |
11013 | 0302-9743 | 0 |
PageRank | References | Authors |
0.34 | 12 | 5 |
Name | Order | Citations | PageRank |
---|---|---|---|
Xiaotian Ma | 1 | 0 | 0.34 |
Mingbo Zhao | 2 | 125 | 10.52 |
Zhao Zhang | 3 | 938 | 65.99 |
Jicong Fan | 4 | 81 | 9.62 |
Choujun Zhan | 5 | 12 | 2.17 |