Implementación de un algoritmo para eliminación de ruido impulsivo en imágenes y análisis comparativo de tiempos de respuesta bajo arquitectura GPU y CPU

Abstract

The research had as a primary objective determine in the field of digital processing images, response times in different architectures (CPU and GPU), using interpolation trough the basis radial functions. To achieve this objective, it start with a previous research about impulsive noise elimination in images, from there its proposed an appropiate algorithm based in a pseudocode solution, to the CPU architecture and GPU architecture. For the GPU architecture is detailed several particularities identified at the implementation phase (using CUDA technology); platform restrictions and implementation work-arounds. As a result of the implementation phase, its proposed a test images, which have noise salt and pepper with different dimensions (width, height), these tests seek to determine response times about noise elimination by the algorithm implemented on the two architectures. In testing the results were analyzed, mainly showing a correct noise elimination in images (reaching up to 55 thousand noisy pixels in a image) at both CPU and GPU architectures, additionally a clearly lower response time (faster in processing) on the CPU Architecture regarding to the GPU Architecture.