A GPU Accelerated Local Polynomial Approximation Algorithm for Efficient Denoising of MR Images

This paper presents a parallelized implementation of the Local Polynomial Approximation algorithm targetted at CUDA-enabled GPU hardware. Although the application area of LPA in the image processing domain is very wide, here the focus is put on magnetic resonance image de-noising. In this case, LPA serves as a pre-processing step in the method based on Shape-Adaptive Discrete Cosine Transform. It is shown, how the designed efficient implementation of LPA substantially reduces the execution time of SA-DCT.