The conventional Bayer filter should have died a long time ago

[Karim D. Ghantous]

Sometimes, people make decisions which are just good enough that their flaws aren't obvious for many years. The USB-A plug is one example of that. The Bayer array is another. (The disappearance of the 3.5mm headphone jack on phones is another).

The Bayer array has 50% of the grid as green. We are told that this mimics human eyes. But, this begs the question. Why should cameras mimic human eyes? I'm not sure why a camera has to share all the characteristics of human vision.

But I think the hidden reason why nobody as changed it is because the green filter lets in more light than the red or blue filters (I read this somewhere but I can't find the source). So, to increase ISO performance, we are stuck with having 50% of the grid as one colour, at the expense of the other two.

The only truly good colour array is one that shares two features: full saturation; and an even spread of R, G and B filters.

The Phase One Tri-Color back, with a fully saturated Bayer array:

(3:30)

 

[Alex Zeder]

I'd personally love to have one of the two green ones replaced with an infrared filter and this then recorded into a separate RAW channel for various use scenarios in post. Much more flexibility than with a full spectrum sensor, but it would for sure also make lens selection more tricky and complex because they all have different qualities in the IR range, even those that match closely in the visible range.

 

[Phil Holland]

Keep in mind the Trichromatic CFA is still a Bayer Pattern. It's just a bit more optimized. Having used the camera, I do like the color off of it. But they aren't the only ones making custom CFA or sensors.

On an absolute side note, I am very curious if Phase One is going to follow up the IQ4 series. It's been a long while. I wouldn't mind a new body at this point.

 

[Les Hillis]

Just day-dreaming here, but with the advances in processing power and machine-learning, it would be interesting to see how accurately colour information could be extrapolated into the extremes of a sensors dynamic range, without having to actually capture colour information at those extremes.

Like, for example, if a sensor had enough colour pixels for capturing an accurate real-life colour reference within a 'normal' resolution and dynamic range, while simultaneously having other pixels capturing higher resolution, monochromatic (or even IR or UV tuned) tonal information of a higher resolution and dynamic range. I wonder how accurately all the data could be combined to create a colour image that would be equivalent in accuracy to it having been shot within the 'normal' range of the sensor (while also taking into account the effect of specific lighting and lens characteristics).

I can see there would still be limits to that method, like if for example a new object that was outside the 'normal' range of capture entered the frame and there was no colour reference for it that could be expanded upon in the first place. Although, even there, pushing the concept even further, that colour information could be guessed or chosen later, like so much of what we see on screen already is.

I know after mentioning it before that some people would be against using that kind of 'artificial' approach to any extent, and I agree there would be cases where any degree of 'artificial' generation of the image would be unacceptable, but if it were developed and applied properly (within measurable and testable limits that allow direct comparisons to be made), I can imagine the results would be practically indistinguishable from 'reality'.

 

[Alex Zeder]

We had a lot of options back when you could remove the OLPF filters totally or change them on RED cameras.

True, but no OLPF can record IR and visible light in two separate streams at the same time.

 

[Zack Birlew]

Keep in mind the Trichromatic CFA is still a Bayer Pattern. It's just a bit more optimized. Having used the camera, I do like the color off of it. But they aren't the only ones making custom CFA or sensors.

On an absolute side note, I am very curious if Phase One is going to follow up the IQ4 series. It's been a long while. I wouldn't mind a new body at this point.

Of course I'd like to see some 4K+ video on these digital backs as well, I mean when you're paying that much for a high megapixel digital back why not add video at this point? There's some interesting use cases for video with medium and large format back adapters. My current DIY rig works with any camera and combines parts of a certain large format camera but while it gets the job done it's definitely not as nice compared to having a streamlined video-capable digital back solution with a basic digital back adapter.

 

[Karim D. Ghantous]

Keep in mind the Trichromatic CFA is still a Bayer Pattern.

We're 50% there. ;-) BTW, do you know if Red's Bayer filters are fully saturated?

 

[Karim D. Ghantous]

Of course I'd like to see some 4K+ video on these digital backs as well, I mean when you're paying that much for a high megapixel digital back why not add video at this point?

In principle, that would be great. But what's the point of 4K from a 100Mpx sensor? It's natively 11K, so if you are using it for video, you want the full 11K. Of course, I don't know if that is possible at this point.

 

[Zack Birlew]

I'm just going off of the Fuji GFX series as they slow walk it to full sensor 8K+ and since digital backs have similar sized or bigger sensors then I'm sure they'd toss a bone for full sensor 4K at least at this point.

 

[Karim D. Ghantous]

I have another solution, for those who don't want a 1:1:1 RGB filter: swap the blue and green filters. Two blue for every green, so it's now 1:1:2. Much better that 1:2:1.

I'm just going off of the Fuji GFX series as they slow walk it to full sensor 8K+ and since digital backs have similar sized or bigger sensors then I'm sure they'd toss a bone for full sensor 4K at least at this point.

There is no point in 4K on such a large sensor. You get all the deficiencies of a large sensor with a resolution that is half that of VV. 12K is at least a 50% boost in resolution (and colour).