2
$\begingroup$

How to detect numbers on the following image

enter image description here

I tried:

  img=Import["http://imgur.com/TDbn7sB.png"]

  set = ImagePartition[img, Scaled[{1/20, 1/5}]] // Flatten

enter image description here

then I tried Classify[] but it doesn't work for large number of data,(greater than 3)

enter image description here

$\endgroup$
  • 4
    $\begingroup$ Have you seen the question Recognizing integers in an image? Perhaps this and this might help as well. $\endgroup$ – C. E. Oct 28 '16 at 9:57
  • 3
    $\begingroup$ I think what that message is trying to tell you is that you have too few training samples. "The number of classes is close to the number of examples" -> you should increase the number of examples. $\endgroup$ – Niki Estner Oct 28 '16 at 10:02
  • 1
    $\begingroup$ @C.E. I don't think the duplicates apply considering vito tries to use Machine Learning to do this $\endgroup$ – Feyre Oct 30 '16 at 16:39
  • $\begingroup$ @Feyre So if I were to ask "how do I binarize an image? I tried using Binarize" that would not be a duplicate of "how do I binarize an image? I tried using ImageFilter," or what? I'm not saying that OP's attempt is a duplicate of another post, I'm saying that the question is a duplicate. If he were to make machine learning a part of the question, then the situation would be different. $\endgroup$ – C. E. Oct 30 '16 at 16:46
  • $\begingroup$ @Feyre The idea with tagging the other question as a duplicate isn't that answers using machine learning aren't welcome, it's that they are just as welcome in that other question as in this one. So the situation becomes ambiguous: if someone wants to post a machine learning-based answer, in which question should he post it? That's the problem we solve by tagging questions as duplicates. $\endgroup$ – C. E. Oct 30 '16 at 16:48