How do I differentiate between a hole and a surface in 2D images?

__

Answer 1

[Image Analyst]

Unfortunately you forgot to attach the image so we'll have to wait for that. If there is a color difference in the image, then yes, you can find the hole. See my 3 color segmentation demos in my File Exchange.

[Aloush]

Yes you are right. I will upload the 2 images soon.

[Image Analyst]

I would try a stdfilt() on one of the color channels. Where it's uniform, like at that round white sticker in the right image, the standard deviation will be small. So use stdfilt() and threshold, then look for round regions of a certain size, as demonstrated in my Image Segmentation Tutorial in my File Exchange

[Aloush]

Thank you for your answer, but some times the details appeared inside the hole also have a small standard deviation. More particularly, assume that there is no texture appeared inside the hole of the left image, the standard deviation will be small also, and may be same as the right one in some situations.

[Image Analyst]

Well, if the hole is fairly smooth and uniform, can you at least assume it will be darker than the top of the cube? If so, try to threshold it.

[Image Analyst]

There is no way from a 2D color image you can determine with certainty that a hole is there or not there with a simple color image. If you can't standardize on your situation, then someone could fool the algorithm by putting a sticker on the box top that had the same color/lightness/pattern as what we'd see by looking through a hole. So to be more robust you'd have to use some kind of range data like you'd get from stereo imaging or Kinect.

[Image Analyst]

I've never played around with structure from motion, or shape from shading, or any things like that. Good luck though. Different viewpoints could help - you can get range information from the stereo pair. What's so interesting about that box anyway? Do you have any real world industrial or scientific problems to solve? Or is this just a simple case to work on before you tackle the real world problem later?