Scale-Space and Edge Detection Using Anisotropic Diffusion 个人总结-CSDN博客

本文链接：https://blog.csdn.net/weixin_53991087/article/details/122284175

论文改进基础：

高斯核+卷积：(48条消息) 高斯模糊的算法（高斯卷积高斯核）_farmwang的专栏-CSDN博客_高斯卷积核https://blog.csdn.net/farmwang/article/details/74452750

其中t表示variance（方差）

As pointed out by Koenderink [111 and Hummel [7],this one parameter family of derived images may equivalently be viewed as the solution of the heat conduction,or diffusion, equation with the initial condition I ( x , y, 0) = I0(x, y ) , the original image

即用扩散方程的解等价于使用高斯去噪的结果

Koenderink motivates the diffusion equation formulation by stating two criteria.
I ) Causality: Any feature at a coarse level of resolution is required to possess a (not necessarily unique)“cause” at a finer level of resolution although the reverseneed not be true. In other words, no spurious detail shouldbe generated when the resolution is diminished.
2) Homogeneity and Isotropy: The blurring is requiredto be space invariant

以上可以由www.zhihu,com/question/24591127得到更加清晰且基础性的解释，同时为后文的理解做好前瞻性的铺垫

The reason for this spatial distortion is quite obviousGaussian blurring does not “respect” the natural boundaries of objects. Suppose we have the picture of a treetop with the sky as background. The Gaussian blurring process would result in the green of the leaves getting
“mixed” with the blue of the sky, long before the treetop emerges as a feature (after the leaves have been blurred together). Fig. 3 shows a sequence of coarser images obtained by Gaussian blurring which illustrates this phenomenon. It may also be noted that the region boundaries
are generally quite diffuse instead of being sharp.
（高斯模糊不尊重物体的自然边界。假设我们有一张以天空为背景的树梢图片。高斯模糊过程会导致树叶的绿色和天空的蓝色混合在一起，在树梢出现之前很久(在树叶被模糊在一起之后)。图3是通过高斯模糊得到的一组较粗的图像序列，说明了这一现象。还可以注意到，区域边界通常是相当扩散，而不是尖锐。）

With this as motivation, we enunciate [18] the criteria which we believe any candidate paradigm for generating multiscale “semantically meaningful” descriptions of images must satisfy.
I ) Causaliry: As pointed out by Witkin and Koenderink. a scale-space representation should have the property that no spurious detail should be generated passing from finer to coarser scales.
2) Immediate Localization: At each resolution, the region boundaries should be sharp and coincide with the semantically meaningful boundaries at that resolution.
3) Piecewise Smoothing: At all scales, intraregion smoothing should occur preferentially over interregion smoothing. In the tree example mentioned earlier, the leaf regions should be collapsed to a treetop before being merged with the sky background.

依照此规则提出了各向异性的diffusion

根据Koenderink的推导，在考虑尺度空间的扩散方程框架中，扩散系数c被假定为一个与空间位置无关的常数。为什么会这样，并没有根本的原因。但这无疑是不合理的，因此P&M将展示如何选择一个合适的c (x, y, t)（即各向异性）使能够满足上一节列出的第二个和第三个标准。此外，这可以在不牺牲因果关系标准的情况下完成。

考虑各向异性扩散方程

用div表示散度算子，用上三角和下三角分别表示梯度算子和拉普拉斯算子，它们与关于空间变量相关。

当c (x, y, t)为常数时，可简化为各向同性热扩散方程

假设在时间(尺度)t时，我们知道适合该尺度的区域边界的位置。我们希望鼓励区域内的平滑，而不是跨边界的平滑。这可以通过在每个区域的内部设置传导系数为1，在边界处设置为0来实现。然后模糊将在每个区域单独进行，区域之间没有相互作用。该地区的边界将保持尖锐。

因此问题的一个关键是要估计清楚边界所在的位置

设E (x, y, t)是这样一个估计:定义在图像上的一个向量值函数，理想情况下应该具有以下性质: 1. E (x, y, t)在每个区域的内部= 0。 2. E (x,y, t) =每个边缘点上的Ke(x,y, t)，其中E为该点上垂直于边缘的单位向量，K为边缘的局部对比度(左右图像强度的差值)。

如果估计E (x, y, t)是可用的，传导系数c (x, y, t)可以被选择为E的大小的函数。根据前面所述的策略g()是一个非负单调递减函数g(0) = 1。这样扩散过程将主要发生在区域的内部，不会影响E值较大的区域边界。