Medical image denoising by generalised Gaussian mixture modelling with edge information

Denoising is a classical challenging problem in medical image processing and understanding. In this study, the authors propose a novel generalised Gaussian mixture model (GGMM) with edge information to denoise medical images. In the first stage, they extend Gaussian mixture model to the GGMM for modelling the noisy medical images and use minimum-mean-square error under the Bayesian framework to derive a non-linear mapping function for processing the noisy images. In the second stage, they refine the results by the kernel density function of the edge information. Experimental results on the Simulated Brain Database and real computed tomography abdomen images demonstrate that GGMM-Edge Information achieves very competitive denoising performance, especially the image grey, visual quality and edge preservation in detail, compared with several state-of-the-art denoising algorithms.