Performance Analysis on Several GPU Architectures of an Algorithm for Noise Removal
Department of Systems and Computing, Instituto Tecnologico de Cd. Guzman, Cd. Guzman, Jalisco, Mexico
The 2012 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA’12), 2012
@article{sanchez2012performance,
title={Performance Analysis on Several GPU Architectures of an Algorithm for Noise Removal},
author={S{‘a}nchez, MG and Vidal, V. and Bataller, J. and Verd{‘u}, G.},
year={2012}
}
In this paper, we present an efficient implementation of parallel algorithms to remove noise in digital images using different Graphics Processing Units (GPUs). The algorithm, based on the concept of peer group, uses a fuzzy metric for finding wrong pixels and the Arithmetic Mean Filter (AMF) to correct it. There are many factors to study in order to get an optimal implementation of an algorithm on a GPU. Our algorithm has been implemented with two different approaches to access the data: Shared and Texture memory. Also, the number of threads and its arrangement have been studied. The test has been conducted on two different cards: Tesla-Fermi and GeForce.
September 1, 2012 by hgpu
