In the next few years, the rate of enhancement in GPUs (Graphics Processing Units) performance is expected to outshine that of CPUs (Central Processing Units), increasing the demand of the GPU as the processor chosen for image processing. In light of tremendous advance in computer vision research of recognition shape domain, we proposed a GPU technology of programming and computing to accelerate the Fourier descriptor technique invariant to color images classification. It is a simple and powerful technique to represent objects based on their shapes. It has attractive properties such as rotational, scale, and translational invariance. Since the heaviest part of Fourier descriptor computing time is the Fast Fourier Transform (FFT), we decided to bring it out on GPU. We used CUDA: Compute Unified Device Architecture, the specific programming language of GPU, and its CUFFT library to accelerate the computation of FFT. To showcase this implementation, we studied the performance of GPU versus a traditional implementation on CPU for single and double precision.
Digital Object Identifier (DOI)
Haythem, Bahri; Fatma, Sayadi; Marwa, Chouchene; Mohamed, Hallek; and Mohamed, Atri
"Accelerating Fourier Descriptor for Image Recognition Using GPU,"
Applied Mathematics & Information Sciences: Vol. 10
, Article 31.
Available at: https://dc.naturalspublishing.com/amis/vol10/iss1/31