Acceleration of super-resolution for multispectral images using self-example learning and sparse representation. (August 2017)

• Record Type: Journal Article
• Authors: Moustafa, Marwa S.; Ebied, Hala M.; Helmy, Ashraf K.; Nazamy, Taymoor M.; Tolba, Mohamed F.
• Is Part Of: Computers & electrical engineering. Volume 62(2017)
• Page Start: 249
• Page End: 265
• Publication Date: 2017-08
• Subjects: Super resolution -- Orthogonal matching pursuit (OMP) -- CUDA -- Sparse representation; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2017.02.012
• Languages: English
• ISSNs: 0045-7906

Abstract

Highlights: Reconstruction high-resolution from low-resolution multispectral image. Perform resolution enhancement based on sparse representation and self-example methods. Implement OMP on graphics processing units (GPGPU) using the Unified Device Architecture (CUDA) to accelerate the computation. Better results compared with the state-of-art bicubic method and some literature super-resolution methods are achieved. Abstract: Super-resolution obtains a new high resolution image from single or multiple low-resolution images for the same scene. Recently compressive sensing has been successfully used in signal recovery. This paper investigates the potential reduction in execution time by selecting tasks that can be parallelized using general purpose computing on graphics processing units (GPGPU) and Compute Unified Device Architecture (CUDA). The self example based super-resolution method via sparse representation and morphological component analysis is proposed for satellite images . Orthogonal Matching Pursuit (OMP) is used in the high resolution image reconstruction phase. The complexity of each module in the OMP algorithm is analyzed and its bottlenecks are identified at the projection module and the least squares module. The projection module is accelerated by adopting a GPU tiled matrix vector multiplication. To speedup the least square module, a GPU implementation of the Jordan matrix inverse algorithm is adopted. Different experiments have been carried out on synthetic and satellite images. Extensive experimental comparisons were conducted with state-of-the-art super-resolution algorithms to validate the effectiveness of the proposed approach. The proposed GPU implementation for OMP is tested on NVS 5200M GPU on Intel ® Core(TM) i7 CPU. The GPU implementation accelerates the speedup compared to the CPU sequential implementation from 20× for small images to more than 40× for large image sizes. Graphical abstract: