Majid Mohrekesh - Department of Electrical and Computer Engineering Isfahan University of Technology, Iran
Shekoofeh Azizi - Department of Electrical and Computer Engineering Isfahan University of Technology, Iran
Shadrokh Samavi - Department of Electrical and Computer Engineering Isfahan University of Technology, Iran

The widespread usage of the contourlettransform(CT) and today’s real-time needs demand fasterexecution of CT. Solutions are available, but due to lack ofportability or computational intensity, they aredisadvantageous in real-time applications. In this paper wetake advantage of modern GPUs for the acceleration purpose.GPU is well-suited to address data-parallel computationapplications such as CT. The convolution part of CT, which isthe most computational intensive step, is reshaped for parallelprocessing. Then the whole transform is transported into GPUto avoid multiple time consuming migrations between the hostand device. Experimental results show that with existing GPUs,CT execution achieves more than 19x speedup as compared toits non-parallel CPU-based method. It takes approximately40ms to compute the transform of a 512×512 image, whichshould be sufficient for real-time applications

contourlet transform; real-time; convolution GPU; CUDA