Parallelization of Digital Image Segmentation in Shared Memory Multicore Systems

Abstract

One of the recent innovations in computer technology is the development of multicore systems that involves two or more co-processing elements to act as independent cores. Computational speed could be quickened by stimulating the multicore processors which constitutes the profound underlying technique for shared memory parallel processing. Shared memory multicore architecture urges innovative research in developing an optimal parallelized version of time consuming scientific applications like segmentation of digital images. Multicore technology could be exploited for the development of parallelized version of image segmentation methods. newlineIn the present scenario, there exist several prominent methods for performing segmentation of digital images with extreme accuracy. But while adapting these methods in real life situations, they will consume huge processing time and worsens the overall performance. This is because image segmentation is computationally intensive and complex in nature. With the advancement in multicore technology, this processing time delay taken by image segmentation methods can be reduced by adapting structured parallelism in shared memory mode. So far there had been not much quantified researches being laid down in parallelizing image segmentation methods in shared memory multicore systems. newlineThis research work focuses on parallelizing three such memory bound image segmentation methods: Gradient Vector Flow (GVF) Active Contour Snake, Maximum-a-Priori Maximum-Likelihood (MAP-ML) method and Jump Flood Cut (JF-Cut) method. Relatively unspecified studies had been laid down with the parallel implementation of GVF Active Contour Snake, MAP-ML and JF-Cut in shared memory multicore systems with Open Message Passing (OpenMP). Even though these methods provide accurate result of the segmented area, they endure huge and complex arithmetic computations there by consuming much of the eand#64256;ective processing time. This perplexity of figuring could be immensely reduced by parallelizing the methods of choice