BROWSE
Detail
Conference Paper
Implementation of Fast Self-Similarity based Super Resolution
Cited - time in 
Share 
Share
- Authors
- Dae Yeol Lee, Jooyoung Lee, Sukhee Cho, Seyoon Jeong, Hui Yong Kim
- Issue Date
- 2016-01
- Citation
- International Workshop on Advanced Image Technology (IWAIT) 2016, pp.1-4
- Language
- English
- Type
- Conference Paper
- Abstract
- The Self-Similarity based Super Resolution (SSSR) method can reconstruct high resolution images without referring to the external database by exploiting high frequency information from the local self-similar region. However, the SSSR process requires computations associated with multi-scale image pyramid and thus, has a high complexity. In this paper, we propose acceleration approaches for major time-consuming operations of the SSSR to speed up the process while minimally affecting the visual quality. First approach is a separable kernel design, where the number of computations in convolution and morphology operations is significantly reduced without affecting the output values. Second approach is a fast similar patch search method which aims to reduce the search candidates while minimally affecting the visual quality of the output. Various fast search algorithms, such as Three Step Search (TSS) and New Three Step Search (NTSS), were applied and their performances with respect to the full search were investigated. Experiment results show that the proposed acceleration approaches are able to reduce the computation complexity of the SSSR significantly without much visual quality degradation.
- KSP Keywords
- Experiment results, Fast Search Algorithms, Frequency information, Full search, High frequency(HF), High resolution images, Kernel design, Morphology operations, Multi-scale, New Three Step Search, Quality degradation