Construcción de un prototipo de agentes para dispositivos móviles que permita segmentar imágenes médicas y realizar procesos paralelos multicore utilizando CPU’s y GPU’s

Abstract

The problem is based on the reduction of times in the execution of processes that have a large amount of calculation, such as the pre-processing of medical images to which segmentation with the Watershed technique is also carried out, and to reduce the over-segmentation that has this technique applied post-processing using Histogram Specification. In order to reduce the calculation time, OpenMP was used. This prototype is developed in Jade-Leap for him to send images from the main container to the mobile device, the mobile application is developed in C++. It was obtained as a result that of the global average of Watershed segments with pre-processing is 53% and 47% for Watershed with post-processing. For parallel processes, 57.8% of the total average execution of the application is for the CPU and 42.2% corresponds to the GPU. For the first scenario, 90% of CPU time was used, the second scenario used 15% of CPU time. That is, there was a 75% improvement comparing the two scenarios.