1
$\begingroup$

This is a simple image classification problem. I want to categorize jigsaw puzzle images to difficulty levels (e.g, easy, medium, hard).

I want to develop an simple algorithm for this classification, which is based on factors like large solid colored areas, repeating patterns on the image (more details). Such classification can be seen in following website. example 1

I want to automate this without much human interventions. Is this possible and whats the best way to approach it. (not worried about performance, simple is good, does not need huge accuracy)

EDIT

I was able to generate the magnitude and phase plots of the 2d FFT. However, I am not clear how to come up with a classification by looking at magnitude plot. Any ideas.

Improve this question
$\endgroup$
3
  • $\begingroup$ I think you do not have a well-defined, calculable metric. $\endgroup$
    – Emre
    Jan 24, 2016 at 19:15
  • 1
    $\begingroup$ The problem here is that you asked a very specific question (large solid color areas) and then came up with a completely different topic (classification of jigsaw puzzles). What you should do before any coding: Think very carefully about what parameters make a jigsaw puzzle hard or easy. Are there correlations between them? And if you came up with, lets say 3 - 5 parameters, then state them here and also give us the reason why you think that those are good parameters. Then it is easier (for us and you) to come up with an idea for an algorithm to quantify those parameters from images. $\endgroup$
    – M529
    Mar 24, 2016 at 12:22
  • $\begingroup$ The 1st link is dead. $\endgroup$
    – Eric Johnson
    Apr 17, 2022 at 14:56

1 Answer 1

Reset to default
1
$\begingroup$

Perhaps you could classify the difficulty by assuming that higher spatial frequency implies higher difficulty. Make sure all the images are the same size, do a 2D FFT, and look at the spatial bandwidth in x and y dimensions, would be my suggestion.

Improve this answer
$\endgroup$
4
  • $\begingroup$ Thank you very much for your answer. I am not an expert in this area. Can you give me some helpful links or general algorithm I could implement and try out? $\endgroup$
    – user2943193
    Sep 26, 2015 at 3:37
  • $\begingroup$ I kind of gave you the algorithm already... do you know what an FFT is? If not, I would start there. $\endgroup$
    – CMDoolittle
    Sep 26, 2015 at 3:46
  • $\begingroup$ I was able to generate the magnitude and phase plots of the 2d FFT. However, I am not clear how to come up with a classification by looking at magnitude plot. Any ideas. Lets say Easy i.telegraph.co.uk/multimedia/archive/02431/… Medium difficult mangomaniafl.net/wpcontent/uploads/2014/06/…, Hard is mytechquest.com/blog/wp-content/uploads/2010/10/… $\endgroup$
    – user2943193
    Sep 27, 2015 at 11:46
  • $\begingroup$ The problem with the pure spatial frequency approach is that it would classify solid colors and very busy noise (of the type in the bottom image of the linked examples) at opposite ends of the easy/hard space when they should be equally hard. I think difficulty is probably a function spatial frequency relative to the piece size with easy being somewhere in the middle. I'm not stating that with great clarity, but the metric isn't terribly clear either... $\endgroup$
    – Omegaman
    Mar 24, 2016 at 19:39

Your Answer

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

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