It is necessary to have denoising pretreatment for images before the calculation of curvature attribute. However, when eliminating noise, conventional methods of image filtering would meanwhile smear the image features such as edges, lines and textures;while the P-M model based on partial differential equations would lead to blocking effects in the smoothing process. In order to solve these problems, this paper proposed an anisotropic filtering pretreatment based on tensor diffusion, which would get abundant information about the local structure of an image by defining the scatter matrix, and then used these structures to control the diffusion process so as to achieve better image filtering. The theoretical analysis and experimental results show that compare to several general algorithms of image filtering, anisotropic filtering can make the result of curvature attribute clearer and higher in quality.%在曲率属性计算之前需要对图像进行去噪预处理,传统的图像滤波方法在去除噪声的同时会破坏边缘、线条、纹理等图像特征,而基于偏微分方程的P-M模型在平滑过程中会出现块效应.针对这些问题,提出了一种基于张量扩散的各向异性滤波的预处理方法.通过定义散布矩阵来获得丰富的图像局部结构信息,然后利用这些结构来控制扩散过程,以便实现图像的更好滤波.理论分析和实验结果表明,相较于一些常规的图像滤波算法,各向异性滤波得到的曲率属性效果更清晰、质量更高.
展开▼
