An Ultra-Fast, Optimized and Massively-Parallelized Curvelet Transform Algorithm on GP-GPUs
Amirkabir University of Technology
Amirkabir University of Technology, 2013
@article{motamedi2013ultra,
title={An Ultra-Fast, Optimized and Massively-Parallelized Curvelet Transform Algorithm on GP-GPUs},
author={Motamedi, Mohammad and Sobhieh, Sima and Motamedi, S Ahmad and Rezaie, Amir Hossein},
year={2013}
}
The Curvelet transform is among one of the most powerful time-frequency representations of an image. However, since it is not a fast algorithm it cannot be employed in most real-time and/or large scale applications. This paper proposes a novel algorithm to speed up the Curvelet transform by both optimizing it for repetitive Curvelet usage and by massive parallelization. In this parallel algorithm we have reconfigured the structure of the Curvelet algorithm in a parallel format suitable for implementation on the GP-GPU. Experimental results prove the proposed algorithm to be ultra-fast without loss of accuracy.
August 5, 2013 by hgpu
