search

fiji / Trainable_Segmentation

Fiji library to perform image segmentation based on the Weka learning schemes
https://imagej.net/Trainable_Weka_Segmentation
GNU General Public License v3.0
106 stars 60 forks source link

wrong formula of orientation feature in documentation? #64

Open AEttinger opened 3 years ago

AEttinger commented 3 years ago

Hello,

I'm currently trying to re-implement some of the Weka features in Python and had trouble to reproduce the orientation feature based on the Hessian matrix. After some digging, I found that the formula on the documentation page [ 1/2 arccos(4b^2 + (a - d)^2) ] does not match the one used in FeatureStack.java [ -0.5 Math.acos((s_xx - s_yy) / Math.sqrt(4.0 s_xy s_xy + (s_xx - s_yy) (s_xx - s_yy))) ]. Using the Feature Stack implementation, my result is now much more comparable to the output of the Weka plugin in Fiji, so I assume that this is being used. I wanted to bring this to your attention, just in case somebody else stumbles over this. It may be good to change it on the documentation page.

Thank you very much. Andreas

ctrueden commented 3 years ago

Thanks, @AEttinger! What do you think, @iarganda?

iarganda commented 3 years ago

Hey @ctrueden and @AEttinger ! Sorry to come late to the conversation. If I'm not mistaken, someone else told me about it before. I tried to update the TWS site, but at the time it was not possible because of the transition of the imagej.net site. Is it possible now?