Best rescale intensity method for preserving image relations


#1

Hi. I want to analyze many images by unsupervised clustering and supervised classification of the CellProfiler output. My images (acquired with a Molecular Devices Image Express instrument) are fairly faint and CellProfiler advises me to rescale intensity, which I have been doing using method (E), entering settings 0, 0.0625, 0, 1, 0, 1. This seems to work OK, but I want to make sure that by rescaling each image independently, I am not introducing intensity variability between images that could throw off subsequent statistical comparison of the images. My question is: what is the best method for rescaling the images without introducing image-to-image variability, so that the intensity, texture and other features can be directly compared between images? (I apologize if this is discussed in Cell Profiler Analyst - I am not using it due to my peculiar data structure)

Many Thanks!

Erik


#2

Hi Erik,

Good question! Fortunately, you have nothing to worry about; RescaleIntensity with option E and the parameters mentioned serve to convert 12-bit images saved in 16-bit format to the correct intensity range, via a simple linear scaling by a constant value. Therefore, the scaling will be identical for every image.

Had you used AA or AE instead of the constants specified, then in that case, you would have been introducing image-to-image variability, which would be problematic for your application. You can take a look at the Help for RescaleIntensity for a few more details.

Regards,
-Mark