基于EM算法参数估计的各向异性扩散超声图像的去噪

目的 提出一种各向异性扩散滤波器的扩散参数选取方法,提高滤波器的灵活性和稳定性.方法 使用二状态的瑞利、高斯混合分布对超声图像灰度分布进行拟合,并采用期望值最大化(expectation maximization,EM)算法实现混合分布的分解;根据分解结果预测图像中斑点噪声均匀分布的区域;通过对均匀区域统计特性的分析获取各向异性扩散的扩散参数.结果 通过与两种改进扩散参数选取的滤波方法对比,基于EM算法的混合分布分解能够准确地估计扩散参数,使滤波结果在噪声消除和边缘保持上达到有效的平衡.结论 基于EM算法参数估计的各向异性扩散是一种有效的超声图像去噪方法.

Speckle Reduction for Ultrasound Images with Anisotropic Diffusion Based on EM Parameter Estimation Algorithm

Objective To improve the flexibility and stability of the filter in speckle reduction of ultrasound images. Methods The gray distribution of an ultrasound image was modeled by a mixture of one Rayleigh and one Gaussian distribution. The mixture distribution was decomposed using the expectation maximization(EM) algorithm. The homogeneous region of the image was then estimated according to the decomposition results of the mixture distribution. Finally, the diffusion threshold was obtained by analyzing the statistical features of the homogeneous region. Results By comparing the results with other two improved parameter estimation methods for the anisotropic diffusion, the method based on the EM algorithm could estimate the diffusion threshold more accurately. The balance between speckle removal and edge preservation could also be obtained in the diffusion result. Conclusion The anisotropic diffusion based on EM parameter estimation is an effective method for suppressing speckle in ultrasound images.