医学超声图像是医生诊断人体组织病变的重要依据, 而医学超声图像中固有的散斑噪声易造成纹理信息的破坏, 影响医生对组织器官的判断, 因此, 医学超声图像的去噪处理倍受关注。针对目前医学超声图像去噪算法无法保持图像纹理这一局限性, 本文提出分数阶微分加权的引导滤波算法。算法首先通过对数变换, 将难以去除的散斑噪声转换为加性噪声; 再结合分数阶微分算法, 根据像素与边缘纹理的相关性设计纹理因子, 并使用该纹理因子改进引导滤波方法; 最后, 通过改进的引导滤波器生成医学超声图像的处理结果。本文对猪胃和猪气管超声图像进行了算法实验, 实验结果表明, 本文算法相较于经典引导滤波算法, 其结构一致性因子提升20.1%, 无参考图像锐化因子提升3.3%, 能够在去除散斑噪声的同时有效保留图像边缘纹理结构, 对于医学超声图像具有良好的适用性。

Medical ultrasound image is an important basis for doctors to diagnose human tissue lesions. The speckle noise inherent in medical ultrasound images is easy to cause the destruction of texture information, which affects the doctor′s judgment on tissues and organs. Therefore, the denoising process of medical ultrasonic images has attracted much attention. In view of the limitation that the current medical ultrasound image denoising algorithm cannot maintain image texture, a fractional differential weighted guided filtering algorithm was proposed. Firstly, the speckle noise was converted into additive noise by logarithmic transformation. Combined with fractional differential algorithm, the texture factor was designed according to the correlation between pixel and edge texture, and the texture factor was used to improve the guided image filtering. Finally, the processing result of the medical ultrasound image was generated by the improved guided image filtering. In this paper, the ultrasound images of pig stomach and pig trachea were tested. Experimental results indicate that compared with the guided image filtering, the proposed method respectively gets 20.1% and 3.3% advancement for Structural Similarity Index Measurement and Cumulative Probability of Blur Detection. It can satisfy the proposed algorithm can effectively preserve the edge texture structure of the image while removing speckle noise.