According to the shortcoming of image preprocessing by Gauss linear filter in coherence enhancing diffusion model, and the eigenvalues of diffusion tensor are not suitable for denoisng in flat area, false edge often induces in plain area, it proposes an anisotropic diffusion denoising method based on morphological operator. This method firstly uses morphological close-open operator instead of Gauss filter to do pretreatment, then structure tensor is designed by second directional derivatives, and eigenvalues of diffusion tensor are devised according to adaptive gradient threshold value. The numerical experiments show that the improved model is capable of removing noise efficiently, preserving the details of image characteristics and eliminating the false edge in plain area commendably.%针对相干增强扩散模型采用高斯线性滤波做图像预处理的不足,及扩散张量特征值的选取不适合平坦区域的去噪,易在平坦区域产生虚假边缘,文中提出了一个基于形态学算子的各向异性扩散去噪方法.该方法首先利用形态学闭开算子代替高斯滤波做预处理,然后结合二阶方向导数设计结构张量,且依据自适应的梯度阈值设计扩散张量的特征值.数值实验结果表明,改进后的方法在有效去除噪声的同时,还能很好地保持图像的细节特征和消除平坦区域的虚假边缘.
展开▼