How we feel depth in 2D pictures

Ever since I got my very first camera, I've noticed that taking pictures is a science, rather than an art. It is related to how we visually perceive information. One thing interesting to me is how we feel depth from 2D pictures. Things seen previously can effect the way we perceive new objects. This problem is related to depth perception. I believe studying how we sense depth from single 2D image can help the research in computer vision and, in a larger extent, artificial intelligence.

Today I just wanna give a simple example using two pictures below of the same Saturn V rocket in Huntsville, Alabama.

Figure A

Figure B
Do you feel that Saturn V in Figure B is "longer"  than it in Figure A?

The difference between the two pictures is on the left side. In Figure B, I kept some banners and building structures. The pictures are 2D and thus there is no way for us to sense the depth by looking at pictures. But those extra stuff on Figure B provides us the feeling of depth according to our experience. We have seen banners on poles along a long road or windows on a long building. So they help us get a feeling about depth and thus we can feel the length of Saturn V better.



If you mask the tandem lights on the left of the rocket, you will feel more difficult to sense the length of the rocket.

I have seen some research using single images (instead of stereo cameras taught in classical textbooks) for depth perception, like this one: http://robotics.stanford.edu/~ang/papers/aaai08-Make3dDepthPerceptionSingleImage.pdf

No comments: