Estimation of dependencies of the running time of the algorithm for recovering defocused images performed on the CPU and GPU
Abstract
Estimation of dependencies of the running time of the algorithm for recovering defocused images performed on the CPU and GPU
Incoming article date: 09.01.2019The subject of research is the problem of choosing the most efficient hardware architecture that implements the algorithm for deconvolution (recovery) of distorted images. The Wiener filter is taken as the deconvolution algorithm under consideration, due to its efficiency, both in terms of image restoration quality and due to acceptable time complexity. The object of study is the process of determining the time complexity of the considered algorithm for the recovery of damaged images when it is executed on a central processing unit (CPU) and a graphics processing unit (GPU). The main functions of blurring and defocusing of images are considered: Gaussian blur, Bokeh effect, Motion blur or motion blur. The research method is based on an experimental assessment of the dependencies of the algorithm operation time based on the Wiener filter, performed on the CPU and GPU, on the dimension of the image being restored. The results of a computational experiment conducted in order to compare the dependencies of the operating time of the Wiener filter, performed on the central processor and the graphic processor, on the size of the image being restored, are presented. Based on the presented results, it was found that when using images whose dimensions do not exceed 1920 * 1080, the Wiener filter is more expedient to implement on the CPU, and when restoring images that are larger than 1920 * 1080 - on the GPU.
Keywords: image deconvolution, Wiener filter, parallelization, recovery of defocused images, CPU and GPU, algorithm optimization