Accelerating fuzzy adaptive anisotropic diffusion on GPU

A new filtering method to remove Rician noise from magnetic resonance images is presented, while harnessing the powerful computational resources of GPUs. In this filter, the direction of diffusion and the characters of different kinds of pixel in noisy MR images are characterized by the eigenvector and eigenvalues of structure tensor. In order to enhance edges, the shock filter based on fuzzy sets is coupled to it. This model can be performed in a memory and computation-efficient way on modern programmable GPUs, which can be regarded as massively parallel coprocessors through NVidia's CUDA compute paradigm. We achieve considerable speedups compared to an optimized GPU implementation and CPU methods for 2D MR image.