Hello everyone,
I am trying to find the most efficient method to predict with accuracy the size of my objects. I thought I could work with the feature “Principal Components of the object” as it seems to give axes of an ellipse fit to the object. However, I absolutely don’t know how a PCA matrix worsk so I don’t know how to use it either, is it possible to use the eigenvalue values I have in my csv file to get the real length of the axes of the ellipse ? 
Hi @Antoine_Cure,
In order to calculate the PCA, all object pixels are taken into account. The eigenvalues will be comparable between similar objects of different sizes and orientations. You can compare those values between objects and they give you an approximation of object sizes. The corresponding eigenvecors give you an idea about the orientation.
If you are after the size of objects you could also look at the “area” (which gives you the number of pixels of the object mask).
I’m sure Fiji should also offer various size measures derived from an object mask image.
Hope this helps
Cheers
Dominik
Hello Dominik,
I am terribly sorry I forgot to respond 
Thanks for your answer it is getting clearer but I am still a bit confused, how is it possible to get an approximation of the object sizes with the PCA ?
I wish you a happy new year and all the best for you ! 
Antoine
Nevermind I found the solution by using the Radii given by Ilastik !