Acelerando o mean-shift com OpenCL

Abstract

In the field of image analysis there are several methods of image segmentation that make up for a more complex sequence of filters and operators applied to a certain set of images. The mean-shift fits in this context, however, their performance falls short in some situations, such as increasing the dataset and its number of dimensions. To fix this problem, we propose a solution that implements a part of the algorithm in OpenCL, looking forward to speed up the process of obtaining the results. The results show a significant increase of the algorithm’s performance, which enables further research using this metodology.