[Python] Adjoint of the Radon transform by non-filtered backprojection in scikit-image

The function iradon of scikit-image implements the filtered backprojection of the Radon transform. By setting filter_name = None one get an unfiltered backprojection. I was expecting that this will be the adjoint map of the Radon transform, but apparently it's not.

A numerical check shows that this is not the case: If A stands for radon and Aᵀ stands for the unfilteren backprojection by iradon, I checked if 〈Ax,y〉 and〈x,Aᵀy〉are equal for randomly chosen x and y of appropriate size - and they are far apart. I suspect that there is a scaling issue and my question is: The the adjoint of the Radon transform called dual transform on Wikipedia implemented somewhere in scikit-image?

Continue reading...