I am currently trying to understand a paper about shadow removal (DESHADOWING OF HIGH SPATIAL RESOLUTION IMAGERY APPLIED TO URBAN AREA DETECTION, Samara Azevedo et al.), but I am currently stuck at this formula:
Equation about anisotropic filtering
The text under the equation says:
where f^(t) corresponds to the scaled version of f, g = g(|∇ G_σ ∗ f^(t)|) represents an edge detection function, and G is the classic Gaussian kernel governed by a tuning parameter σ.
I am not sure if it is better suited at the StackOverflow site, as it is a computer vision related task, but if I am wrong just correct me. So now my questions:
- Could somebody guess in which way f_(t) is scaled? As f_(t) - f doesn't make any sense if the images are not the same size.
- Could somebody explain the ∇ to me? Usually, it denotes an edge detection algorithm (e.g. Laplacian), but as it is not described I am not 100% sure. Especially here: g = g(|∇ G_σ ∗ f^(t)|) I am not sure what the ∇ does, as calculating edges on a kernel is not useful (I think).
I know this is not a lot of information and i cannot post the paper here directly, but as the results of this paper look promising I would like to understand it, though it seems like there is some information missing (maybe the authors consider it common knowledge)
Thanks!