1
$\begingroup$

Given that there is an image with objects in it, is there any method in computer vision to determine the z-ordering of objects. For an ideal image of three circles overlapping, I can determine that it is three circles overlapping and determine which is the closes to me in this image.

Therefore for any given image, is there any method in computer vision to determine the z-ordering of segmented parts?

To clarify, I don't mean for only clean geometric objects such as circles, I mean for any objects which have been segmented.

I have a particular application of this in mind, I want to be able to reverse engineer images into layers, then I can pass the image split into layers into photoshop to speed up some of what we do practically with photoshop which is splitting images into layers. Also layers which are partially hidden can then be filled in automatically by the computer, layers can be moved on top of each other.

enter image description here

$\endgroup$
  • 1
    $\begingroup$ Huh... I can only see one circle, and two lunes placed right next to it in a chain... $\endgroup$ – John Dvorak Nov 30 '14 at 11:30
  • $\begingroup$ I mean, how are you going to handle ambiguous inputs? Even if you can distinguish which parts of image belong to which object, it doesn't mean there is an unambiguous ordering to them. For what it's worth, I could be looking at a distant lighthouse through a carefully cut out cardboard image of a clear blue sky. And even if relative occlusion is unambiguous, it may or may not depict a valid partial ordering. A could occlude B while being occluded by C, which is occluded by A. Or, I can have two objects, each occluding the other (think: two chain links) $\endgroup$ – John Dvorak Nov 30 '14 at 11:39
  • $\begingroup$ Regarding your use case, even if you manage to z-order stuff correctly (and cut up layers where impossible), filling in occluded stuff from the context is hard. $\endgroup$ – John Dvorak Nov 30 '14 at 11:41
  • $\begingroup$ For your first "ambiguous inputs", my inputs are not that complex, most cases are simple. Filling in a image area from content around it has been done before, anyway its a different problem from this problem. $\endgroup$ – Phil Nov 30 '14 at 12:13
  • $\begingroup$ "simple" is a relative term. Telling if a photo is a photo of a bird sounds simple, but computers still suck at it. Even Google, which has overabundance of test cases, can fail. $\endgroup$ – John Dvorak Nov 30 '14 at 12:15

Your Answer

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

Browse other questions tagged or ask your own question.