1
$\begingroup$

Assume you have 77777 pictures that are mostly noise. They are about magnetic anisotropy.

There is one picture among these pictures which has high contrast difference between two large areas (some sort of magnetic gap because of magnetic anisotropy). One area is gray and the other area is dark.

The pictures are mostly like this (grey or dark)

enter image description here

and

enter image description here

There are some dirty things in some pictures which are just rubbish on the lens. So only big differences are essential. Example what the contrast difference should look like

enter image description here

These pictures are about magnetic anisotrophy with Fe and Boron.

How can you find the distinction among these 77777 pictures?
I have went through all of these pictures but my method is errorprone, since I just browse them through.

How can you assist browsing the find the big and large contrast difference in pictures?

Any tool is ok for me: Matlab, Python, ..., pseudo-code too.

$\endgroup$
  • $\begingroup$ It it not clear to me which pictures out of the group should be marked as different. Can you show some more examples? $\endgroup$ – Maurits Mar 21 '15 at 16:42
  • $\begingroup$ @Maurits I added an example picture to the body. There should be a clear contrast difference between areas. $\endgroup$ – Léo Léopold Hertz 준영 Mar 21 '15 at 16:56
1
$\begingroup$

Implement a low pass 2D filter that will reduce greatly the noise. Following an example that you could complete yourself. You could subtract the filter bias to keep an average of 0 "addition" by the filter, controlling the dynamic range.

Example of filter

         1/16  1/8   1/4  1/8   1/16

    1/16  1/8   1/4   1/2  1/4   1/8     1/16

1/16  1/8   1/4   1/2   1    1/2   1/4     1/8     1/16

1/8   1/4   1/2   1     2    1     1/2     1/4     1/8

1/16  1/8   1/4   1/2   1    1/2   1/4     1/8     1/16

   1/16  1/8   1/4   1/2  1/4   1/8     1/16

        1/16  1/8   1/4  1/8   1/16

You might be able to clearly see the desired image.

Than run an Edge a 2D edge detector which will enhance the location of the step. You could try running two 1D filters: -1 1 in horizontal and vertical direction. If you know the relative "black"/"white" areas you could select which to use of the optional -1 1 or 1 -1.

You might be able to use frequency domain filtering for the low pass 2D filter. If you are not familiar with this type of challenges and this is not exercise you need to get a consultant to help you or study the subject in more depth.

Depending of the contrast of the "step" you might be able to detect automatically the "line" created by the edge detector.

$\endgroup$
  • $\begingroup$ Can you propose some low pass 2D filter and some Edge 2D edge detector? Any ready implementation for Matlab or any other tool? $\endgroup$ – Léo Léopold Hertz 준영 Mar 21 '15 at 11:16
  • $\begingroup$ Thank you for your edit! Is there are ready solution in Matlab? $\endgroup$ – Léo Léopold Hertz 준영 Mar 21 '15 at 17:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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