How to analyze "regularity" of an image?

Tags: #<Tag:0x00007fd54092d228>


Hi! I’m wondering how I can analyze the regularity of this type of image.
This is an example of the result of an experiment. I’d like to describe numerically the homogeneity/heterogeneity of this image. An increase of bright and dark areas, white spikes at the margins and low homogeneity of the grey color in this case, mean low quality of the result.
Is there an index that describe such a features?
What procedure/plugin/analysis would you suggest me to quantify the degree of heterogeneity of this image? May somebody help me?
I read about kurtosis but I’m not sure that is the right choice.
Thank you for your help!


Good day!

Your sample image appears to suffer strongly from JPG-artifacts. Could you please post an uncompressed raw image in TIF- or PNG-format.

Is the sample image really in the best original spatial resolution. Presently, the spatial resolution is rather low …




As a possible global measure of homogeneity you could try the decline of the raw Fourier power spectrum of the image as a function of the radial spatial frequency, i.e. integrated over concentric circles. The latter can be achieved by using the ImageJ-plugin “Radial Profile”.

A related approach would start with the auto-correlation function of the image instead of the Fourier power spectrum.




Thank you very much!
I have an uncompressed version of the image (the forum does not let me upload it) but the quality is still low because the acquisition module of the microscope has its limits.

I’ll study what you suggested :wink:


If you’re looking for a first “cheap & fast” solution, then try the Standard-Deviation divided by the Mean.

getRawStatistics( N, mean, min, max, std );
print( std / mean );

Paste the above two lines of macro code to an empty macro window (Plugins >> New >> Macro) and run it.




Good idea. May be enough for this purpose. I’ll give it a try with some images clearly different each other.
I want to understand also the Radial profile procedure. It sounds interesting.
I’ll let you know if it works.
Thanks a lot


Here are three results achieved with Fourier power-spectral analysis:

  1. Your sample image

  2. White noise

  3. Constant gray value

All spectra were normalized to the DC-value, i.e. their values at the spectral origin were one. The displayed graphs show cropped signals with maximum displayed value 10^-5.