An Adaptive Scheme for X-ray Medical Image Denoising using Artificial Neural Networks and Additive White Gaussian Noise Level Estimation in SVD Domain

Abstract

Medical image denoising is an important and one of the most challenging fields of biomedical image processing. The presence of noise reduces the visual quality of medical images and impairs the ability to perform accurate diagnosis and treatment. The principal aim of denoising is to improve the perceived quality of images, remove the undesired noise, while preserving the diagnostically relevant information. This paper proposes an adaptive scheme for denoising of X-ray medical images. The proposed method adopts the use of multiple multilayer perceptrons to perform image denoising. Each multilayer perceptron is trained to perform image denoising at a specific noise level. The proposed method relies on the singular value decomposition of images to estimate the level of additive white Gaussian noise that is present in images. In an attempt to optimize the performance of the proposed method, the paper investigates how the choice of image segmentation block size and various options for the multilayer perceptron architecture affect the ability of artificial neural networks to perform image denoising at various noise levels. The performances of the proposed image denoising method is evaluated on a database X-ray images. The experimental results demonstrate that compared to a single MLP based approach to image denoising, the proposed image denoising scheme improves and provides a more consistent image denoising performance across noise levels.