jf_web
Welcome!
Everything on this website was built and programmed by me, some of it even in languages and libraries I understand! However, it would not have been possible without the immense body of literature and work made freely available by countless people all over the world. In that spirit, the entire code for this website is freely available here.
See below some choice interactive course notes – check out the Teaching tab for more.
Speckle as a wave interference phenomenon
Let’s look an illustrative real-world manifestation of this conceptual framework. SAR images are subject to speckle, a type of image distortion or noise which affects pixels in a stochastic, multiplicative way. Why does this occur?
Consider that the returned wave associated with a given pixel likely isn’t the result of a single interation with a single pointwise object. If multiple reflectors (or similarly, a continuous reflector) lie within the pixel, then the returned wave is the sum of multiple individual waves.
Let’s imagine we have a pixel with two identical point reflectors lying within it. How does the returned wave change as a function of where those reflectors are located? Move the blue and red targets in the pixel below:
Distance from target one to the satellite: m
Distance from target two to the satellite: m
Here we’ve considered a SAR satellite 600km above Earth, emitting 50cm wavelength radiation with a pixel resolution of 3m.
The distance from each target to the satellite affects where in their cycle each wave is re-intercepted, and thus the difference determines whether those returned waves interfere constructively or deconstructively. Thus a random distribution of targets creates a random difference in distance to the satellite, creating a random level of interference in each pixel. This means the intensity of each pixel is randomly (though not necessarily uniformly) multiplied by a value from 0 (perfect deconstructive interference) to 1 (perfect constructive interference). This is the origin of speckle, and reason why it manifests as multiplicative noise (as opposed to additive).
Let’s now consider two pixels: one with more, stronger scatterers and one with fewer, weaker scatterers – perhaps pixel one belongs to some dense shrubland, while the other is a freshly plowed field. The distribution of scatterers is likely quite random – what happens to the brightness of these pixels under different configurations?
Clicking through a few dozen random configurations of each pixel, we can see in the developing histogram of their brightnesses that while the pixel with more strongly reflecting targets is on average brighter, there are many times when the pixel with weaker targets actually gives a higher return. This is due to the random level of interference we’ve called speckle. We can exploit the fact that the average value remains higher to identify likely high-return areas in speckle filtering. As an aside, notice that the distributions of brightness appear relatively Gaussian – this is a lovely invocation of the Central Limit Theorem.