The use of GPUs in image processing

Author

Frasheri, Mirgita ; Cico, Betim

Author_Institution

Comput. Eng. Dept., Polytech. Univ. of Tirana, Tirana, Albania

fYear

2013

fDate

15-20 June 2013

Firstpage

124

Lastpage

127

Abstract

The analysis of climatic parameters, vegetation, humidity and pollution in the domain of time and space is done by processing a series of images of a geographic area taken by the satellite at certain times [1]. These images are subject to several computing schemes, with the aim of evaluating spatial and temporal variations of the mentioned parameters. One of the programs used to manipulate the images is the CHERS, which is completed in the framework of the FP7 project SEE-GRID-SCI [2]. This program calculates polynomial trend in time for pixels of ordered sets of images. In this paper we have considered the parallelization of the CHERS algorithm. The parallelization technique implemented in this study is Cuda, which is used to program multicore NVIDIA GPUs. It is observed a decrease in user and system time proportional to the number of active threads. Also, CPU percentage falls to a minimum of 69%.

Keywords

geophysical image processing; graphics processing units; multiprocessing systems; parallel architectures; polynomials; remote sensing; CHERS algorithm; CUDA; FP7 SEE-GRID-SCI project; climatic parameter analysis; geographic area; humidity; image processing; parallelization technique; pollution; polynomial; program multicore NVIDIA GPU; spatial variation evaluation; temporal variation evaluation; vegetation; Approximation methods; Graphics processing units; Kernel; CUDA; NVIDIA; geographic area; image processing; multicore;

fLanguage

English

Publisher

ieee

Conference_Titel

Embedded Computing (MECO), 2013 2nd Mediterranean Conference on

Conference_Location

Budva

ISSN

1800-993X

Type

conf

DOI

10.1109/MECO.2013.6601335

Filename

6601335