Removing Reflections with Independent Component Analysis
Taking a photo of something that is behind a glossy surface (like through a window) usually results in an unwanted reflection. One way to remedy this is to use a polarized lens. This will remove a large portion of the relfection from the image. However, it is not 100% effective and also, if we don't have access to a polarized lens, we still want good photos!
We can think of the image behind the glossy surface and the image formed by the reflection over the surface to be a linear combination of two independent images. We know previously that convolving independently identically distributed random variables produces a Gaussian (central limit theorem).
From "Separating Reflections and Lighting Using Independent Components Analysis" by Farid, H. and Adelson, E.
If we want to separate independent components in the image, and we know that convolved independent components make a somewhat Gaussian product,
we can separate the components by maximizing "non-gaussian-ness." One such measure of "non-gaussian-ness" is negative entropy.
