基于梯度的自适应非局部均值超声图像去噪方法

Gradient-Based Adaptive Nonlocal Mean Ultrasound Image Denoising Method

摘要: 为解决医学超声图像中存在的斑点噪声问题，从而影响了为临床医生提供准确病理诊断的能力。本文提出了一种基于梯度的自适应非局部均值算法，用于去除医学超声图像中的斑点噪声。该算法利用中值滤波预处理噪声图像，并且计算噪声图像的梯度，再利用一种新的非线性二值化方法对梯度图像进行处理，以获得图像自适应的衰减参数，由此确定相似度权重。最后利用到非局部均值算法中得到去噪图像。大量实验表明，在三种不同噪声强度的仿真图像上与几种经典去噪算法的效果相比，本文提出算法的峰值信噪比(PSNR)平均提高了13%，结构相似性(SSIM)平均提高了15%；在三张真实临床医学图像上与几种经典去噪算法的效果相比，本文提出算法的等效外观数(ENL)分别提高了158%、26%和88%，对比度噪声比(CNR)分别提高了15%、25%和23%。相较于几种经典的去噪算法，本文提出的算法能更有效地抑制斑点噪声、保留小结构并增强图像对比度。

Abstract: In order to solve the problem of speckle noise in medical ultrasound images, it affects the ability to provide accurate pathological diagnosis for clinicians. In this paper, an adaptive gradient-based nonlocal homogenization algorithm is proposed for de-speckling noise in Medical ultra-sound images. The algorithm preprocesses the noisy image using Median filtering and calculates the gradient of the noisy image and then processes the gradient image using a new nonlinear binarization method to obtain the image adaptive attenuation parameters from which the similarity weights are determined. Finally the denoised image is obtained by utilizing to the non-local mean algorithm. Numerous experiments show that PSNR of the proposed algorithm in this paper is improved by an average of 13% and SSIM is improved by an average of 15% when com- pared with the effects of several classical denoising algorithms on three simulated images with different noise intensities; ENL of the proposed algorithm in this paper is improved by 158%, 26%, and 88%, and CNR is improved by 15%, 25%, and 23% when compared with the effects of several classical denoising algorithms on three real clinical Medical images, respectively. Compared to several classical denoising algorithms, the algorithm proposed in this paper is more effective in suppressing speckle noise, preserving small structures and enhancing image contrast.