I have a rather difficult image processing image. I would like to rank order a set of images I have by their sharpness. The issue is the images themselves are not of the exact same thing. Usual methods I see rely that the underlying blur-free image is identical, but in my case this is not true.

Some methods I have tried:

  • integration of a band of spatial frequencies (i.e., 2D Fourier transform of image)
  • average of gradient
  • wavelet transform
  • eigenvalue decomposition

None of these have worked so far. They all work, in principle, if the image underneath is identical. Does anyone have some ideas how they would approach this problem?

  • 3
    $\begingroup$ I don’t think this is possible. If you have an all-white image, is this a sharp picture of a featureless white wall, or a really blurry picture of something else? $\endgroup$ Oct 30, 2021 at 0:34
  • $\begingroup$ @CrisLuengo, It depends on the distribution of the images under test. $\endgroup$
    – Royi
    Oct 31, 2021 at 6:35
  • 1
    $\begingroup$ Can we assume natural images or are we talking on a different certain type of images? $\endgroup$
    – Royi
    Oct 31, 2021 at 6:35

1 Answer 1


You may use the approach in the paper A No Reference Perceptual Blur Metric (Available on ResearchGate).

The idea is very simple and described in Measuring Sharpness / Contrast for Auto Focus and Interpreting results of Sobel Edge Detection.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.