LarsonSekanina Convolution cosmetic bug

[RBA]

The parameters Radial Increment and Angular Increment show two sliders each. One of them is always inactive, but it's still there annoying the heck out of me 

There should only be one slider per option, me thinks...

This is with PI SE, 01.05.09.0561 eng (x86_64) on Win7x64

[Carlos Milovic]

Just from what I remember (don't have PI here...). Those sliders work for different filter scales. The lower one, for small scale (0.1 to 10px), and the other, for larger ones. As soon as you reach the upper or lower boundary, the other slider should activate.

[RBA]

Just from what I remember (don't have PI here...). Those sliders work for different filter scales. The lower one, for small scale (0.1 to 10px), and the other, for larger ones. As soon as you reach the upper or lower boundary, the other slider should activate.

You are correct, sir. The bug was in my brain then. It's fixed now 

[Juan Conejero]

Just for the record, LarsonSekanina needs a strong revamping. Its interface works, but it should be updated to meet current PCL standards. On the other hand, LS is an algorithm that I personally don't like, but this is only a personal feeling. Everything that can be "done" with LS can be done with wavelets, except the artifacts

[Carlos Milovic]

Yes, but we should redefine wavelets in a radial coordinate system.

[Juan Conejero]

You can transform the image from Cartesian to polar coordinates, perform a wavelet transform, then the inverse transform from polar to Cartesian. We have all the required tools in PCL: PolarTransform and ATrousWaveletTransform. Sounds well doesn't it

[Carlos Milovic]

Hmmm... not convinced at all just for one problem: edge artifacts. I believe that the black pixels from "outside of boundaries" coordinate values in a R/Theta image will create a lot of them. Also, a common approach for the upper and lower boundaries (i.e. theta = 0 or 2Pi) may not work. It should be like a fourier function, not mirroing.