1
$\begingroup$

Can an algorithm be derived to classify digital images according to how they were generated, i.e. the classes would be:

  • Real-world photograph captured with an actual camera.
  • A Computer-Generated realistic scene.
  • A hand-drawn image or a painting.

Could perhaps a Machine Learning approach, geared towards analyzing statistics of the signal provide sufficient insight on the problem ? In that case, is there any known implementation of such algorithm publicly available ?

$\endgroup$
  • $\begingroup$ Putting online and use the google image advanced search ? :-p ( beside kidding, they might have publish (white) papers on that ). $\endgroup$ – Fabrice NEYRET Mar 12 '16 at 0:10
  • $\begingroup$ Questions asking us to recommend or find a book, tool, software, tutorial or other off-site resource are off-topic as they tend to attract opinionated answers and spam. Instead, describe your problem or need and the steps, if any, you've taken to solve it. $\endgroup$ – trichoplax Mar 12 '16 at 11:25
  • $\begingroup$ As it stands, this question does not specify what types of images will need to be considered, so I am closing as unclear. The question may be reopened if it can be edited to clarify what types of non-photographic images will be presented, and to request an algorithm rather than an off site resource recommendation. $\endgroup$ – trichoplax Mar 12 '16 at 11:28
  • $\begingroup$ See also What topics can I ask about here? $\endgroup$ – trichoplax Mar 12 '16 at 11:29
  • $\begingroup$ I would expect the histogram of real-world image to be a lot more noisy and chaotic that CG generated or CG-painted ( If image textures are not used ). Of course, it's a pretty coarse criterion. $\endgroup$ – Fabrice NEYRET Sep 13 at 11:02
0
$\begingroup$

Difficulty depends on what you really want to classify, i.e. what exactly your data base of images is made ok. For simple hand-drawn or computer art, I would think of just looking the color histogram (but for complex art with gradients; but still these gradients are likely to induce a very smooth colormap).

At the other extreme, distinguishing realistic computer graphics image from reality is probably impossible.

| improve this answer | |
$\endgroup$

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