Variational Method for Image Denoising by Distributed Genetic Algorithms on GRID Environment

The aim of this paper is to present a novel distributed genetic algorithm architecture implemented on grid computing by using the G-Lite middleware developed in the EGEE project. Genetic algorithms are known for their capability to solve a wide range of optimization problems and one of the most relevant features of GAs is their structural parallelism that fits well the intrinsically distributed grid architecture. The proposed architecture is based on different specialized autonomous entities able to interact in order to carry out a global optimization task. The interaction is based on exchange of knowledge on the problem and solutions. In this way the main problem can be solved by using many cooperative small entities that can be classified into different specialized families that cover only one aspect of the global problem. The topology is based on archipelagos of islands that interact by chromosomes migrating with a user-definable strategy. Grid has been mainly used in the high performance computing area. The properties of the proposed GAs architecture and its related computing properties have great potential in solving big instances of optimization problems. Furthermore this implementation (distributed genetic algorithms with grid computing) is suitable to solve time consuming problems reducing by executing different instances on many virtual organizations (VOs) according to the grid philosophy. The proposed parallel algorithm has been tested on denoising problems applied to image processing which are known to be time consuming. The paper reports some results about the time performance compared to traditional denoising filter algorithms