Pixel shift

[ts1]

Here's my belated entry - shift the sensor by a pixel every other frame. For stationary objects, you'd then be able to combine successive frames to get a green sample at each pixel location.

This will help recover most of the detail lost by traditional demosaicing algorithms. Many popular demosaicing algorithms first reconstruct green, and then use that to extract red and blue values. The green channel is also responsible for most of the luma information in such schemes.

A good demosaicer gets you about 70% of information on each axes for a total of 6 megapixels in the Mysterium's case. This is slightly below the 8 megapixel that 4K cinema is capable of. Boosting the all-important green resolution to nearly 100% will help surpass the 4K cinema specs with room to crop.

This scheme does not help moving objects, but then fine detail is not perceptible there anyway. Especially at 24 fps with all that is done to keep the flicker down.

 

[ts1]

Someone seems to have implemented a variant of this scheme already. See "Jitter Camera" at http://www1.cs.columbia.edu/CAVE/projects/jitter_camera/jitter_camera.php

My technique is simpler in that it requires one actuator instead of 2, and is more tolerant of motion (it requires 2 successive frames instead of 4). On the other hand it only doubles the megapixels instead of quadrupling it.

My scheme also works with the demosaicing algorithm and not on top of it. It is also likely to generate fewer artefacts (in still areas) since it directly samples green at all pixel locations.

In summary, it is a cheaper, simpler and a more robust way to get half the resolution of the Jitter Camera.

-Tripurari