You'd be surprised. Depending on the resolution of the photo, I might be able to get a low-res image of what's behind the door through the light leaks in its cracks. The more I know about the camera you used, the specific setting you took that photo in, the better chance I have at making a good guess of what's behind the door. Or behind the camera. These are the "assumptions" I'm talking about. External knowledge applied to resolve ambiguities.
For simpler examples, compare:
- GPS signal power at the receiver is under the thermal noise floor, yet they can still be recovered if you know exactly where to look.
- Nonlinear image transformations like blurs can still be inverted if you know or can guess the transform and exact coefficients used. Which is why, to redact something, you really want to paint over it with uniform color, instead of blurring it.
For simpler examples, compare:
- GPS signal power at the receiver is under the thermal noise floor, yet they can still be recovered if you know exactly where to look.
- Nonlinear image transformations like blurs can still be inverted if you know or can guess the transform and exact coefficients used. Which is why, to redact something, you really want to paint over it with uniform color, instead of blurring it.
Or just a lot of what photogrammetry is about.