2 ms·
I believe it would a homography transform, which is less constrained than an affine transform, as it does not preserve parallel lines/angles. Instead it maps li
by enchiridion 6y ago
I believe it would a homography transform, which is less constrained than an affine transform, as it does not preserve parallel lines/angles. Instead it maps lines to lines as a way to adjust perspective.
https://en.wikipedia.org/wiki/Homography https://en.wikipedia.org/wiki/Homography
- hasmanean 6y agoCan you just tilt the camera sensor to get a transformed image? I remember reading that box cameras allowed you to tilt the film image plane to match a skewed object plane in closeups.
- enchiridion 6y agoYou probably could, but seems a bit excessive to use hardware to solve this problem.
- hasmanean 6y agoWell it removes the problem entirely from the software domain. If it’s a single purpose imaging device with a fixed geometry like OP suggested, it’s the natural solution.
- enchiridion 6y agoTo calculate the angle of the camera you'd still have to find the corners of the paper using some simple computer vision. You could also just use the corners to compute the homograph transform. Really depends on the constraints too. If you're going for archival quality scans then maybe it would be better to do it with hardware. At that point I'd just get a way higher res camera than needed, transform and down sample as needed. Would still probably be cheaper than a gimbal.