Parallelization of Skeletalization of Retinal Background Images in CUDA Architecture

Abstract

This work presents a comparative analysis of the response time for the skeletonization of images, using two versions of the Zhang-Suen algorithm: a sequential mono-processed version andan parallel multi-processed version using the graphics processing unit. The parallel computing platform chosen was CUDA. The skeletonization applications developed is aimed towards processing retinal images, whose characteristics are extracted of blood vessels to assist medical diagnosis, and thus the response time of the system is paramount. Tests were performed on the DRIVE public retinal images database, and showed that the parallel version of the algorithm was, on average, more than 31 times faster than the sequential version.