# Anisotropic diffusion

I am working with speckle noise reduction in ultrasound images. At present I am learning about anisotropic diffusion.I will explain first what I have understood about anisotropic diffusion. I learnt the difference between isotropic and anisotropic diffusion using the diagram below: In case of isotropic diffusion diffusion of the pixel value takes place all across the image.This result in blurring of edges whereas in case of anisotropic diffusion smoothing or diffusion is carried out depending on the image edges and their direction. It smoothens homogeneous image regions but retains image edges. (I know that the tool that is used to find the strength of the edges and the direction is gradient but I need some more explanation on what is gradient? How we calculate gradient? Please explain me in a lucid manner I have zero knowledge about gradient. Please explain me with easy explanation) Based on this gradient value we perform smoothing. If the magnitude of the gradient is small then smoothing is done. If the magnitude of the gradient is large then smoothing is stopped. This graph illustrates the relationship between anisotropic diffusion and gradient magnitude.In the figure the orginal signal is noisy but 2 major peaks are still visible.Applying the diffusison process to the original signal and using the gradient magnitude to attenuate the diffusion process at places where the diffusion process at places where the edges is strong produce a signal in which the SNR is high.The diffusion at the edges of the signals is attenuated by using a function of the gradient magnitude.(This gradient magnitude equation is what I am unable to understand please see the following equation and explain me in detail. I don't Know why they use divergence and gradient )    ( I couldn't able to understand these equations. please explain the equations in relation with the theory above)

Let's proceed step-by-step:

a gradient relate to a measure of the derivate, that is, the coefficient of variation. If you think of a measure of position, the derivate is the speed. Now, we do not have a single point, but a field of pixel with values. Now think of it as a landscape, the luminance as the height and the gradient as the slope. It has both a magnitude (how steep it is) and a direction.

In the formulation that you cite, the slope is computed thanks to the divergence operator. In this case it equals to the sum of slope in the directions of the reference frame (that is, in x and y). It is used to allow a more general formulation in non-euclidean spaces.

This gradient magnitude equation is what I am unable to understand please see the following equation and explain me in detail. I don't Know why they use divergence and gradient

the basic idea behind this equation is to control diffusion by the contours:

• diffusion helps to smooth out data by averaging locally. one drawback is that if you do it too long, you lose all information.
• contours is the abstract concept of a boundary. Think of it for instance as the contour of a object in an image or the transient in the square wave that you show. To detect the position of these contours, you may use the gradient, and specifically the amplitude of the gradient (how steep the slope is)
• now comes the heuristic of anisotropic diffusion: if you diffuse, but that you control the speed of this diffusion by the borders, then, you obtain a better result.

These equations just lay down these concepts quantitatively.

• What is divergent ? How it is related to gradient in image processing? Apr 4, 2014 at 9:34
• edited my answer to incorporate that. hope that fits better to your question! Apr 6, 2014 at 19:48