Speckle Noise Reduction of Medical Imaging via Logistic Density in Redundant Wavelet Domain

Abstract

In the digital world, artificial intelligence tools and machine learning algorithms are widely applied in analysis of medical images for identifying diseases and make diagnoses; for example, to make recognition and classification. Speckle noises affect all medical imaging systems. Therefore, reduction in corrupting speckle noises is very important, since it deteriorates the quality of the medical images and makes tasks such as recognition and classification difficult. Most existing denoising algorithms have been developed for the additive white Gaussian noise (AWGN). However, AWGN is not a speckle noise. Therefore, this work presents a novel speckle noise removal algorithm within the framework of Bayesian estimation and wavelet analysis. This research focuses on noise reduction by the Bayesian with wavelet-based method because it provides good efficiency in noise reduction and spends short time in processing. The subband decomposition of a logarithmically transformed image is best described by a family of heavy-tailed densities such as Logistic distribution. Then, this research proposes the maximum a posteriori (MAP) estimator assuming Logistic random vectors for each parent-child wavelet co-efficient of noise-free log-transformed data and log-normal density for speckle noises. Moreover, a redundant wavelet transform, i.e., the cycle-spinning method, is applied in our proposed methods. In our experiments, our proposed methods give promising denoising results.