[qume]

Cameras only see one colour per pixel (i.e. Bayer colour filter array). Because our eyes are not so acute at resolving colour compared to luminescence this tradeoff works well.

But in this context I'm not sure at a pixel level how much 'extra' information is added by _removing_ red/green from the blue pixels, green/blue from the red etc.

In other words, if the image came from a digital camera, to actually stuff more information into the same number of pixels, colour may or may not help. In any case working with the raw CFA bayer source image would almost certainly be beneficial over interpreting the image after it has been converted to a normal RGB image (losing information in the process)

[JdeBP]

Information is not encoded in pixels, it is encoded in so-called modules, which are more than 1 pixel in either dimension (and square when printed).

[neetdeth]

This is an interesting point, but it assumes perfect optics. If the image is slightly out of focus, I think three lower resolution images in different spectral regions is a big win.

[qume]

I think it depends on the context - if you're designing the whole system from bar code to camera, I'm not sure that the color will help. But in that case you can assume close to perfect optics and I'm not sure if you would find an improvement or the opposite over a monochrome setup with better SNR.

On the other hand if you're stuck with things like mobile phone cameras, and can only control the bar code side, then Id imagine you'd see some improvements as you say.

An interesting middle ground would be if you could get the raw image from the sensor before a generic debayer algorithm gets applied.

Good luck, keep up the good work... i've worked commercially on optical barcode systems and it's a super interesting topic (tho looks like my post has been downvoted out of existence so maybe you wont see this).

[neetdeth]

I don't want to represent myself as one of the inventors! I just saw this on HN, skimmed the technical report, and found a bunch of comments on HN that could be answered with my cursory knowledge thus gained.

What I can offer is that it is now possible to get memory buffers containing "raw" sensor data from mobile phones. I've only done it on iPhone so far, but the "camera2" API on Android looks to support this as well. It only works in single-shot photo mode - I suspect there isn't the bandwidth to do 30 fps streaming, and they rely on the ISP for debayering and color space conversion in video mode.

iPhones seem to have negligible chromatic aberration in their raw output, weirdly, so that isn't a blocker for full sensor resolution grayscale imaging. Someone could exploit this to write the world's greatest mobile phone QR reader app.