CS: Bad Image reconstruction, 3D cameras, hacks and calibration.

If there is anything annoying from the videos of one hundred naked citizens being imaged by X-ray scanners at airports, it's the fact that after all the money spent, the reconstruction solvers are just plain terrible!

I just came across the following article entitled Single-shot depth-camera design by Yung-Lin Chen, Chuan-Chung Chang, Ludovic Angot, Chir-Weei Chang and Chung-Hao Tien starts with: Adding an axially symmetric phase-coded mask and applying an appropriate metric to characterize image blur can provide distance estimates for individual images using a single camera.

The concept is not new ( see here, here or here as start) but it looks like that unlike previous attempts depth is well calibrated.

In the end though, while getting depth from one camera is very nice, how is this feat going to compete with cheap two cameras being hacked by everybody as is the case of the numerous Kinect Hacks.Of specific interest are the attempts of understanding how depth is being computed in these entries:

• Kinect Hacking 101: Hack a Powershot A540 for Infrared Sensitivity.
• Kinect Hacking 102: Foot-Pedal Operated Camera.
• Kinect Hacking 103: Looking at Kinect IR Patterns.

All this to say that the most important part of transfer function engineering or reverse engineering revolves around the difficult business of calibration.