Non-local Means Based Medical Ultrasound Images And Image Sequences Denoising

Medical ultrasound imaging has been widely applied in clinic diagnoses, therapy and surgical navigation because of its non-invasiveness, real-time visualization, low cost and convenience. Because of its coherent imaging mechanism, the speckle noise in the ultrasound image may cover the details of the image and reduce the contrast ratio of human normal to diseased tissue. The negative factors make it difficult of edge detection, image segmentation and image registration. Therefore, the study of effective denoising algorithm for ultrasound images has important theoretical and practical significance.The speckle noise of US image is essentially multiplicative noise and tissue-dependent, therefore, US image denoising is a complex and difficult work. The classical medical US image filters such as SRAD, SBF, filering based on PDE and other methods cannot effectively filter out speckle noise. The non-local means filter is a novel method, which is introduced recently. However, this method has its drawback and needs improvement.This dissertation studies the non-local means algorithm and applies it for US image despeckling and the main resarches are concentrated on the following aspects:First, a new non-local method was proposed by introducing the geometric moments. This method can get better despeckling performance. Geometric moments change the way of similarity measure, and more candidate similarity windows can be picked out for image restoring. Comparing different orders of the geometric moments, the better order was found for obtaining good results.Secondly, a fast blockwise non-local filtering based on zero-order moment was developed. It shortens the run time and is applied to the despecking of US image.Finally, the local and non-local means based mixed filtering were addressed. This method can effectively used in US image despeckling and video denoising.Simulation and experimental results of the actual image evaluation parameters (PSNR, MSSIM, ENL) indicate that the proposed algorithm can effectively filter out the speckle noiseand protect the image edge information and details.