基于多尺度非线性扩散的超声图像斑点噪声去除算法

针对现有去噪算法可能造成超声图像细节模糊甚至丢失的问题,本文提出基于多尺度非线性扩散（multiscale nonlinear diffusion,MSND）的超声图像去噪模型.该模型结合冗余拉普拉斯塔形数据分解和非线性扩散的优点,利用冗余拉普拉斯塔形数据分解将图像分解为等大小的空间-频率子带,综合各子带的特征得到图像边缘和细节的精细表示,然后根据所得的综合特征指导各子带图像的非线性扩散.实验结果表明本文算法在去除噪声的同时能有效地保留和增强边界与细节.

An Ultrasound Image Denoising Algorithm Based on Multiscale Nonlinear Diffusion

In order to solve the problem that current denoising methods might smooth or lose the significant details of image, a new ultrasound image denoising model based on multiscale nonlinear diffusion （MSND） was proposed. This model combined the strongpoint of redundant Laplace pyramid decomposition and nonlinear diffusion. Redundant Laplace pyramid decomposition was utilized to decompose the image into isometric space-frequency sub-images. The edges and details of the image were well expressed by synthesizing the characteristics of all the sub-images. Then the nonlinear diffusion processing was carried out according to the synthesized characteristics in each sub-image. The experimental results showed that the proposed algorithm was capable of filtering speckle noise efficiently and preserving edges and details of images.