I have been using the scikit-image library for a segmentation task in my master thesis. More specifically I have used the regionprops function for calculation of major and minor axis length of objects. I was wondering why exactly, the eigenvalues are multiplied by 4, before returning the axis length? I realize that you multiply by 2 to get the full length, and I think you multiply by 2 again because of the standard deviation, but I am unsure about this.