Paralelização de algoritmo de processamento de imagens digitais

Abstract

This dissertation discuss the related issues of the parallelization of sequencial applications parallelization used to reduce the runtime of experiments and scientific simulations. Much this applications has been written on FORTRAN or C languages, in a period which has not the software and hardware facilities that we have in the present. A large slice of this applications requires a long runtime. The various ciencitifc areas can take the advantages of the parallelization and execution of this algorithms on a cluster, that can be acquired for a low cost than supercomputers. It's possible to process a larger data and execute more tasks that, previously, was impracticable in a unique execution because the computacional cost. In this context it self insert in this present work, that has the main objective the purpose and parallelization of 2 parallel models on MPI and HLRC platforms, of a image processing algorithm applied to geographic images for analysis of the fragmentation index. That algorithm uses the convolution technique to enhance the borders and texture standards allowing analyse the fragmentation index of image. The execution of sequencial version of the convolution algorithm on a unique image can takes up to 25 minutes of processing, however, a set of larger images can be take a lot days. The purposed parallel models are based on PCAM methodology. From this purposed models, 4 parallel versions were developed, that were executed in a less execution time in about 73,9%, 85,5%, 86% e 92,2% than the sequencial algorithm version.