Friedrich-Alexander-Universität Erlangen-Nürnberg, Germany (FAU) proposes a low fill factor pixels pseudo-randomly rotated in the pixel array to improve resolution by a factor of 4 and avoid alias artifacts. Microlens usage and complicated signal routing in the pixel array are not mentioned:
Isn't this a really old idea? I remember proposing it many years ago, but I believe I found I wasn't the first.
ReplyDeleteOne big challenge would be PRNU resulting from variations in photodiode capacitance.
I fail to understand how this should work? Especially when colors (Bayer pattern) come into play... Would they generate e.g. a 3840 x 2160 image from a 1920 x 1080 Bayer pattern?
DeleteI believe it only adds a layer of random blur over the CFA to reduce moiré, at the cost of details. The example shows less moiré and less details/more random smudge.
DeleteWeird example photo, the red zoom in looks better on the left but the blue zoom in looks better on the right.
ReplyDeleteI'm with you there. The "improved image" seems to suffer from some unfortunate JPEG/DCT compression artifacts which make it look a lot worse.
DeleteWith lens CRA change and Bayer pattern, it seems to be something working only in a perfect world.
ReplyDeleteBy itself it is an interesting idea, because MTF and alias depend on the shape of the light sensitive part of the pixel (although a micro-lens can kill this effect). But I think that the cross talk between the pixels is going to be non-uniform across the array and that will generate some other beautiful issues. But a very useful topic to understand the ins and outs of MTF, alias and cross-talk.
ReplyDeleteI did a lot of this in my PhD thesis back in 1993: https://www.era.lib.ed.ac.uk/handle/1842/14043
ReplyDeleteHave a look at page 200 onwards for extensions of the idea, just cite me :-)