PARALLEL PARTICLE SWARM OPTIMIZATION FOR IMAGE SEGMENTATION

One of the problems faced with Particle Swarm Optimization (PSO) is that this method is simply time consuming. It is so, especially when it deals with a problem that needs a lot of particles to represent. This paper tries to compare the speed of PSO run at parallel mode to ordinary one. The testing applies an example of an image segmentation to demonstrate the PSO method to find best clusters of image segmentation. Best clustering is determined by viewing it as it is an optimization problem in finding the minimum error of the clustering. The PSO process, especially the iteration; the one that is the most time consuming; can be fastened by the usage of the parallel property of the PSO. We use NVIDIA CUDA for parallelizing the computation occurred in each particle. The results show that PSO run 170% faster when it used Graphic Processing Unit (GPU) in parallel mode other than that used CPU alone, for number of particle=100. This speed is growing as the number of particle gets higher.