GPU based implementation of spatial fuzzy c-means algorithm for image segmentation

In this paper a meaningful parallel implementation of spatial fuzzy c-means (SFCM) is presented. It has an advantage of being a powerful tool of classical fuzzy c-means. The great effort made to come up with this work is to reduce significantly its complexity and time execution simultaneously. This technique is inspired by the technological progress of GPUs hardware. The studies we have conducted are very relevant especially for large images. We have implemented this parallel algorithm using Compute Unified Device Architecture (CUDA) on different NVidia GPU cards. The numerical results in terms of execution time demonstrate a gain up to 80x for GTX 580 versus the sequential implementation.