New FujiFilm X-Pro1 mirrorless

Greg Smith said:
Those two sensor diagrams are reversed, right? The middle one is the standard Bayer.

- Greg
Click to expand...

Yes, fixed, thank you.

Rick Burnett said:
I don't see how they can call it random since it is still a pattern, just not a symmetric pattern. Nevertheless, I look forward to seeing what it looks like. I think it's a good idea. And yes, your diagrams are backwards! :)
Click to expand...

Which is why its called "pseudo-random". The concept is to reduce a fixed repeating patterns as much as possible compared to traditional bayer that would reduce undesirable aliasing and moire.

Concept in practice (from link above):

diagramprbf.png

diagramprbf2.png
 
kittykatro said:
The concept is intriguing, but why is the random pattern sensor still having 50% of the pixels green? Wouldn't it make more sense to have each of the three colors represented equally?
In a 6x6 pattern you could have for example 12 of the each red,green and blue pixels.
Click to expand...

The human eye is most sensitive to 555 nm wavelength of light; which is yellow-green in color. The human retina has photoreceptor cone cells (S, M, L), these cones aren't one particular color like a bayer filter of a sensor but rather captures a designated spectrum of light. So there are overlapping spectrum that each cone can capture, the M & L cones have intersecting spectrum at around 555 nms, meaning that we see that color near that spectrum very well. This is why 'caution' signs are that yellow-green color, because it pops out at us. Doubling green photosites is designed to simulate this effect to give a more natural image.
 
Film noise is gorgeous. CCD and CMOS noise is painful.

I miss grain. This could be a nice development.
 
Steve Castle said:
The human eye is most sensitive to 555 nm wavelength of light; which is yellow-green in color. The human retina has photoreceptor cone cells (S, M, L), these cones aren't one particular color like a bayer filter of a sensor but rather captures a designated spectrum of light. So there are overlapping spectrum that each cone can capture, the M & L cones have intersecting spectrum at around 555 nms, meaning that we see that color near that spectrum very well. This is why 'caution' signs are that yellow-green color, because it pops out at us. Doubling green photosites is designed to simulate this effect to give a more natural image.
Click to expand...

Yeah...but what's the point of making image sensors with more green pixels, when we already perceive this color very well?
Wouldn't this make red and blue appeared more washed out?
1)because we are less sensitive to these colors
2)because there are less red and blue pixels in a sensor.

Wouldn't the common sense dictate that it should be the other way around: less green pixels and more red and blue?
 
j said:
Film noise is gorgeous. CCD and CMOS noise is painful.

I miss grain. This could be a nice development.
Click to expand...

Hard to say it would be the same kind of look, though. Film grain is random from frame to frame and therefore appears to dance around; this would be a fixed pattern behind the image, although random-ish in the larger sense. It might be annoying like looking at the image projected on a groundglass or something. I'd need to see this in action to make up my mind.

- Greg
 
kittykatro said:
Yeah...but what's the point of making image sensors with more green pixels, when we already perceive this color very well?
Wouldn't this make red and blue appeared more washed out?
1)because we are less sensitive to these colors
2)because there are less red and blue pixels in a sensor.

Wouldn't the common sense dictate that it should be the other way around: less green pixels and more red and blue?
Click to expand...

Because we perceive that color so well, sampling it less means that we'd recognize problems with the image in the greens much more readily. You are mistaking bit depth of the color with the spatial resolution of the color. The red and blue have same same bit depth as green, and spatial frequency doesn't change that. What does change is the color space. The green has twice the resolution coming in as blue or red. That extra resolution has extra detail in it that we do see because we can detect small changes in the green more than the blue and red. The extra detail in the red and blue, because our eyes are not as sensitive, would get lost because we couldn't see the differences in the color shades. That's the theory behind it and it is well understood.
 
Steve Castle said:
Yes, fixed, thank you.



Which is why its called "pseudo-random". The concept is to reduce a fixed repeating patterns as much as possible compared to traditional bayer that would reduce undesirable aliasing and moire.

Concept in practice (from link above):

diagramprbf.png

diagramprbf2.png
Click to expand...

Ahh, so on this new sensor, are they actually doing the pattern like above? Or are they repeating the 6x6 over and over. Definitely two different things. What you have above has randomness to the pattern.

The interesting thing is how the grain affects (b). The extra spacing between the pixels in that image give it more texture, albiet noise texture, where the example above is smoother, but with fixed pattern structures.

Thanks for sharing all this. I really like the technical information. Is there a white paper you can point me to on this with more details and example?
 
Rick Burnett said:
Ahh, so on this new sensor, are they actually doing the pattern like above? Or are they repeating the 6x6 over and over. Definitely two different things. What you have above has randomness to the pattern.

The interesting thing is how the grain affects (b). The extra spacing between the pixels in that image give it more texture, albiet noise texture, where the example above is smoother, but with fixed pattern structures.
Click to expand...



There are many ways to randomize pixels, usually in a pseudo-random bayer the green pixels are always in the same pattern and only the red and blue are randomized.

Given the available information I can merely guess what Fuji is doing, but I'm assuming that its a repeating 6x6 pattern.

Keep in mind, we are moving from a 4 pixel set to a 36 pixel set, this in itself dramatically increases randomization. But looking at the actual pattern its a bit more elegant than that. We have a central 2x2 green pixel set, and by adding the top/bottom and left/right we have two more 2x2 green pixel sets in each of the 36 pixel (6x6) CFA. More importantly, each of the red and blue photosites are offset so that the 4 green pixels are adjacent to different colors.

This is important in that in a traditional 2x2 bayer the cross-talk between each of the R,G,B photosites are always constant. This Fuji sensor is using 2x2 green blocks so that each of the 4 pixels has different cross-talk values (and we have three of these in each 36 pixel set). What this means is that further randomization can be added during demosaicing due to this particular layout.

Rick Burnett said:
Thanks for sharing all this. I really like the technical information. Is there a white paper you can point me to on this with more details and example?
Click to expand...

The whitepaper which I took those images from is linked in my previous post.
 
Instruction manual says "focus, exposure and white balance are adjusted automatically throughout recording".

Does that mean this can't be turned off? What's the point of the movie mode then? Even if I am shooting little Johnny in the backyard, these are undesirable features.
 
> Recent Zacuto shootout contradicts this statement. Film grain looked quite bad, especially with high iso film.

Oh OK, I must be completely wrong then. I should have never looked with my eyes, I should have just read what the internets said. Never mind.
 
I think film grain looks good, but not high ISO film myself. It has ALWAYS bothered me. But, like anything, these are just personal tastes.
 
Rick Burnett said:
Because we perceive that color so well, sampling it less means that we'd recognize problems with the image in the greens much more readily. You are mistaking bit depth of the color with the spatial resolution of the color. The red and blue have same same bit depth as green, and spatial frequency doesn't change that. What does change is the color space. The green has twice the resolution coming in as blue or red. That extra resolution has extra detail in it that we do see because we can detect small changes in the green more than the blue and red. The extra detail in the red and blue, because our eyes are not as sensitive, would get lost because we couldn't see the differences in the color shades. That's the theory behind it and it is well understood.
Click to expand...

This is an excellent explanation. Thanks.

It can be thought of as a form of up-front compression based of color frequency instead of spatial detail. And it works best when you're not going to be color correcting the image very much. But the more you move the image's colors around to make extreme changes, the more this approach literally beings to fall apart before your eyes.
 
BobbyMurcerFan said:
Zacuto's most recent shootout or a hundred years of filmmaking history... which to believe?
Click to expand...

If you shoot high iso film at low light, you would realise that it is quite horrid. Only recent CMOS advances have made it really possible to shoot at lower light levels. There is a reason film productions carry a lot of light.

So saying that cmos-sensors automatically have horrible looking grain is quite wrong.
 
We may be talking about two different things. CMOS can have a cleaner image at low light because of the high ISO - we can agree.

I think though, given an equal amount of grain in each image, the film grain will look better than digital grain (noise).

Some of the most beautiful footage I ever shot was 500 asa 16mm bleach bypass. Grainy? Oh yeah. You could just watch the silver crystals dance around and it was gorgeous.

For me the three big differences between film and video are Resolution (about equal now), Latitude (catching up), and grain. At this point I prefer how film grain resolves fine detail in an organic fashion with each frame a random pattern. To me it just looks good. It looks so good that I even like the grain and noise at times. There are never any times I like the grain and noise in video. Not yet. That's why this thread is exciting to me.
 
You must log in or register to reply here.
Back
Top