针对传统各项异性扩散模型容易产生板块分区, 模糊图像细节等问题,提出一种改进的各项异性滤波算法。该方法引用双曲正切函数构造扩散模型的扩散系数, 避免了传统各项异性扩散模型在均匀区域产生的板块分区问题; 并通过使用衰减因子提高在非均匀区域的扩散敏感性, 能够更好地保留图像的边缘细节信息; 同时引入相对平滑增量, 自适应监控图像的滤波程度, 并自动终止迭代过程。仿真实验表明, 提出的滤波方法能够有效滤除超声图像斑点噪声, 并消除传统各项异性扩散模型产生的板块分区问题, 提高了对图像细节信息的保留能力, 并增强了与原图像的结构相似度, 是一种有效的斑点噪声降噪方法。

In order to eliminate the block effect and the blurring details of images caused by the traditional speckle reducing anisotropic diffusion, an improved anisotropic diffusion algorithm was proposed, in which a new diffusion coefficient constituted by the hyperbolic tangent function was used to replace the original diffusion coefficient, thus can wipe off the block effect in the uniform region of images. A damping factor was used to guide the speed of attenuation in the no-uniform area, thus reserving the details and the weak edges of images. In the meanwhile, the relative smooth increment was introduced to monitor the degree of filtering automatically, and could stop the iteration process of the partial differential equation adaptively. The experiments show that the proposed method can not only filter the speckle noises of the images effectively, but also eliminate the block effect problem lead by the traditional anisotropic diffusion. In addition, it can improve the ability of preserving the detail informations of images and enhance the structural similarity between the filtered images and the original images.