Xtrans vs Foveon vs Bayer vs Monochrome for BW ..... ???

Truman Prevatt · Dec 30, 2019

Tom Schum said:
Truman Prevatt said:

So my answer would be for the highest quality a monochrome sensor. That followed by the Foveon X3 sensor since in both cases there is no interpolation to produce the final image. After that may be the Sigma Quattro with pixel binning to eliminate any interpolation. Next is the XTrans and Bayer CFA's. I expect either the XTrans or Bayer are a wash compared to one another.

Click to expand...

Once you downsize Quattro to 1/4 original pixels, you might as well use the Fuji version, also downsized to 1/4 original pixels. Quattro will be 5mp and Fuji will be 6mp.

There seems to be some enduring controversy about whether or not Sigma Foveon Quattro imagery actually is interpolated or not. Chrominance might be interpolated, but I don't think Luminance is (just my opinion of course). It is certainly a different universe compared with any CFA such as Fuji.

And it goes without saying that there are many threads about all this in the Sigma forum here on DPR. Most of them get pretty technical too.

Sigma DP3M (50mm fixed lens) vs Fuji X-E1 with 23mm F1.4. Admittedly, this is not Quattro to CFA, but it's close.

Yea I saw the food fights of the Q vs M sensor. However, the lower two levels of the Q sensor have 1/4 the number of sensors as the upper level and the output number of pixels is the same as the upper level. So there had to be some interpolation. I know the Foveon purist did not take to it well. But given Foveon and afterwards Sigma tried and failed to trim the noise in the Foveon X3 - they had little choice.

Rule of thumb the "real resolution" of the Bayer is 1/2 the number of detectors. That makes since since the highest resolution channel is 1/2 the number of detectors. You can't create information when none is there. The XTrans might be a little more subtle but in reality I doubt if it is any one or any less than the Bayer.

According to the Peterson-Middleton theorem (generation of the Shannon-Nyquist Theorem to two dimensions) is defined by the largest wave-number that can be resolve which is defined by the sampling of the channel with the most detectors (green in case of CFA and XTrans). That's it - that's the limit no matter how one holds their mouth what kind of snake-oil the manufactures try to sell - you can invent information above this resolution.

I had the SD9 for awhile since I was intrigued by the Foveon sensor. I can tell you that the clarity, detail and perceived resolution of a 3 MP X3 sensor was far superior to anything else out at the time - which was in the 10-12 MP CFA. Make no mistake the Sigma blew the Nikon D200 and DX2 away as far as clarity and detail.

Foveon just could never tame the noise and increase the ISO and dynamic range on the sensor. It is pretty clear why now but its a shame. If I could get a 16 MP Foveon X3 (Sigma starting lying at some point and today they would call it a 48 MP sensor) with the same noise and DR and ISO performance of the latest APSC Fuji is using - I would have one in a nanosecond.

--
"If you learn only methods, you’ll be tied to your methods, but if you learn principles you can devise your own methods." Ralph Waldo Emerson
___
Truman
www.pbase.com/tprevatt

dmaclau · Dec 31, 2019

Tom Schum said:
dmaclau said:

Tom Schum said:

PS when doing monochrome you can get great acutance by downsizing the image to 1/4 total pixels using the "Nearest Neighbor" algorithm. This works great with my Fuji images.

Click to expand...

Nearest Neighbor algorithm? That's new to me, what is it?

Being a big B&W fan and having shot 3 of the 4 mentioned this is an interesting thread for me. If there's another tool I can use I'm very interested

Click to expand...

I use Photoshop Elements. I recently upgraded from Elements 13 to Elements 2020 but they are both almost the same in every respect, including but not limited to resizing algorithm choices.

When you go to resize an image, you have a choice of algorithms. One of the choices is "Nearest Neighbor". Most people use the default, Bicubic, and never even drop down the menu to choose other methods. I'm guessing you have this choice with Lightroom or Photoshop but again, most users never pull down the menu.

This "Nearest Neighbor" algorithm seems to work best when dividing by a square. That is, you can go to 1/4 the original pixels by dividing the original pixels in the horizontal direction by 2, or dividing the original pixels in the vertical direction by 2. Or you can divide by 3 or 4 or 5. So you end up with 1/4, 1/9, 1/16, or 1/25 of the original pixels.

When you divide by two, you get a final image that is 1/4 the original pixels. Each of the pixels in the final image is the average of a group of four pixels in the original image. This basically eliminates the interpolation artifacts of CFA (Color Filter Array) sensors. Yes, there are differing opinions about how far you really have to go, with some saying that Foveon single pixel resolution is equivalent to Bayer that has been downsized to 1/2 original pixels, but try it with any of your existing images in your own library and you will get single pixel resolution in the final image that is equal to or better than Foveon.

With late production Fujis, you get 24mp resolution. When you downsize you get 6mp, and the premier Foveon was the SD15 with 5mp. So they are competitive.

Late production Foveon is the Quattro variant: 20mp resolution for luminance (top layer or "Blue" layer), 5mp "Green" (middle layer), and 5mp "Red" (bottom layer). So, if you use the top "Blue" layer you get 20mp monochrome. This 20mp is pure, non-interpolated image.

Thank you, Tom.

curiouspeter · Dec 31, 2019

Truman Prevatt said:
Foveon just could never tame the noise and increase the ISO and dynamic range on the sensor. It is pretty clear why now but its a shame. If I could get a 16 MP Foveon X3 (Sigma starting lying at some point and today they would call it a 48 MP sensor) with the same noise and DR and ISO performance of the latest APSC Fuji is using - I would have one in a nanosecond.

A Bayer sensor with twice the area and twice the pixels should match the resolution and clarity of an X3.

OTOH, sometimes I find the Foveon X3 files to be too sharp at the pixel level and it can be a challenge to deal with aliasing artifacts.

Personally, I find the 26 MP X-Trans to be a good compromise.

dmaclau · Jan 1, 2020

Truman Prevatt said:
Tom Schum said:

Truman Prevatt said:

So my answer would be for the highest quality a monochrome sensor. That followed by the Foveon X3 sensor since in both cases there is no interpolation to produce the final image. After that may be the Sigma Quattro with pixel binning to eliminate any interpolation. Next is the XTrans and Bayer CFA's. I expect either the XTrans or Bayer are a wash compared to one another.

Click to expand...

Once you downsize Quattro to 1/4 original pixels, you might as well use the Fuji version, also downsized to 1/4 original pixels. Quattro will be 5mp and Fuji will be 6mp.

There seems to be some enduring controversy about whether or not Sigma Foveon Quattro imagery actually is interpolated or not. Chrominance might be interpolated, but I don't think Luminance is (just my opinion of course). It is certainly a different universe compared with any CFA such as Fuji.

And it goes without saying that there are many threads about all this in the Sigma forum here on DPR. Most of them get pretty technical too.

Sigma DP3M (50mm fixed lens) vs Fuji X-E1 with 23mm F1.4. Admittedly, this is not Quattro to CFA, but it's close.

Click to expand...

Yea I saw the food fights of the Q vs M sensor. However, the lower two levels of the Q sensor have 1/4 the number of sensors as the upper level and the output number of pixels is the same as the upper level. So there had to be some interpolation. I know the Foveon purist did not take to it well. But given Foveon and afterwards Sigma tried and failed to trim the noise in the Foveon X3 - they had little choice.

Rule of thumb the "real resolution" of the Bayer is 1/2 the number of detectors. That makes since since the highest resolution channel is 1/2 the number of detectors. You can't create information when none is there. The XTrans might be a little more subtle but in reality I doubt if it is any one or any less than the Bayer.

According to the Peterson-Middleton theorem (generation of the Shannon-Nyquist Theorem to two dimensions) is defined by the largest wave-number that can be resolve which is defined by the sampling of the channel with the most detectors (green in case of CFA and XTrans). That's it - that's the limit no matter how one holds their mouth what kind of snake-oil the manufactures try to sell - you can invent information above this resolution.

I had the SD9 for awhile since I was intrigued by the Foveon sensor. I can tell you that the clarity, detail and perceived resolution of a 3 MP X3 sensor was far superior to anything else out at the time - which was in the 10-12 MP CFA. Make no mistake the Sigma blew the Nikon D200 and DX2 away as far as clarity and detail.

Foveon just could never tame the noise and increase the ISO and dynamic range on the sensor. It is pretty clear why now but its a shame. If I could get a 16 MP Foveon X3 (Sigma starting lying at some point and today they would call it a 48 MP sensor) with the same noise and DR and ISO performance of the latest APSC Fuji is using - I would have one in a nanosecond.

I queue'd up for the original Sigma DP1, was astounded by the color and also the B&W. I likely would still use it except for me it was egg shell fragile. My DP3M is more robust and produces even more incredible imagery. "Noise" banding and overwhelmingly cumbersome post processing sours me on these tools.

I never had any success with Blue Channel only B&W. With filters or without my results never quite met my expectations.

I find my Fuji cameras to be capable in B&W but believe that a B&W sensor camera would be the best image producer.

The differences between Bayer and X Trans sensors are filter arrangement, the sensor concept is the same. Differences in results would be more indicative of photographer or situations than sensor arrangement I believe.

--
absolute power corrupts...absolutely

Truman Prevatt · Jan 1, 2020

curiouspeter said:
curiouspeter said:

Foveon just could never tame the noise and increase the ISO and dynamic range on the sensor. It is pretty clear why now but its a shame. If I could get a 16 MP Foveon X3 (Sigma starting lying at some point and today they would call it a 48 MP sensor) with the same noise and DR and ISO performance of the latest APSC Fuji is using - I would have one in a nanosecond.

Click to expand...

A Bayer sensor with twice the area and twice the pixels should match the resolution and clarity of an X3.

But an X3 with twice the area and the 1/2 the pixels of a Bayer of the same area would produce better images because of no interpolation. The problem is Foveon could never tame the noise in the lower detectors nor could they tame the conversion from their native color space (which was not RGB) to the RGB color space.
I would sure like Fuji to release a special version of the XPro3 without a CFA - simply a monochrome DNG output. I would buy it in a nanosecond. I still have all my color filters from my film days. :-D

curiouspeter · Jan 1, 2020

Truman Prevatt said:
curiouspeter said:

Truman Prevatt said:

Foveon just could never tame the noise and increase the ISO and dynamic range on the sensor. It is pretty clear why now but its a shame. If I could get a 16 MP Foveon X3 (Sigma starting lying at some point and today they would call it a 48 MP sensor) with the same noise and DR and ISO performance of the latest APSC Fuji is using - I would have one in a nanosecond.

Click to expand...

A Bayer sensor with twice the area and twice the pixels should match the resolution and clarity of an X3.

Click to expand...

But an X3 with twice the area and the 1/2 the pixels of a Bayer of the same area would produce better images because of no interpolation. The problem is Foveon could never tame the noise in the lower detectors nor could they tame the conversion from their native color space (which was not RGB) to the RGB color space.

I would sure like Fuji to release a special version of the XPro3 without a CFA - simply a monochrome DNG output. I would buy it in a nanosecond. I still have all my color filters from my film days. :-D

mrb375 · Jan 1, 2020

TRIODEROB said:
think about which of the sensor gets you closest to the real silver halide film look.

we have

1) Xtrans which is found in the Fuji cameras

2) Foveon which is found in Sigma cameras

3) Bayer which is the normal type digital camera tech

4) dedicated Monochrom which is found in the Leica

I think I have gotten great shots with the Xtrans in B&W.. some in 16mp, too!

--
https://www.digitalandfilm.com/

Tom Schum · Jan 2, 2020

Truman Prevatt said:
I would sure like Fuji to release a special version of the XPro3 without a CFA - simply a monochrome DNG output. I would buy it in a nanosecond. I still have all my color filters from my film days. :-D

I would very much like to see this too. It would be a low-production model, and of course there is the need for fundamentally different infrastructure inside the firmware to process images without interpolation. The whole thing might be more than Fuji wants to take on for only a handful of sales, I guess. At least the hardware implementation could be the same as in the X-T3 or X-Pro3, with only a different sensor package mounted.

I'd love to see it though!

Smuj · Jan 2, 2020

Tom Schum said:
dmaclau said:

Tom Schum said:

PS when doing monochrome you can get great acutance by downsizing the image to 1/4 total pixels using the "Nearest Neighbor" algorithm. This works great with my Fuji images.

Click to expand...

Nearest Neighbor algorithm? That's new to me, what is it?

Being a big B&W fan and having shot 3 of the 4 mentioned this is an interesting thread for me. If there's another tool I can use I'm very interested

Click to expand...

I use Photoshop Elements. I recently upgraded from Elements 13 to Elements 2020 but they are both almost the same in every respect, including but not limited to resizing algorithm choices.

When you go to resize an image, you have a choice of algorithms. One of the choices is "Nearest Neighbor". Most people use the default, Bicubic, and never even drop down the menu to choose other methods. I'm guessing you have this choice with Lightroom or Photoshop but again, most users never pull down the menu.

This "Nearest Neighbor" algorithm seems to work best when dividing by a square. That is, you can go to 1/4 the original pixels by dividing the original pixels in the horizontal direction by 2, or dividing the original pixels in the vertical direction by 2. Or you can divide by 3 or 4 or 5. So you end up with 1/4, 1/9, 1/16, or 1/25 of the original pixels.

When you divide by two, you get a final image that is 1/4 the original pixels. Each of the pixels in the final image is the average of a group of four pixels in the original image. This basically eliminates the interpolation artifacts of CFA (Color Filter Array) sensors. Yes, there are differing opinions about how far you really have to go, with some saying that Foveon single pixel resolution is equivalent to Bayer that has been downsized to 1/2 original pixels, but try it with any of your existing images in your own library and you will get single pixel resolution in the final image that is equal to or better than Foveon.

With late production Fujis, you get 24mp resolution. When you downsize you get 6mp, and the premier Foveon was the SD15 with 5mp. So they are competitive.

Late production Foveon is the Quattro variant: 20mp resolution for luminance (top layer or "Blue" layer), 5mp "Green" (middle layer), and 5mp "Red" (bottom layer). So, if you use the top "Blue" layer you get 20mp monochrome. This 20mp is pure, non-interpolated image.

Thanks very much for that, I’ll definitely give this a go. Love B&W images and I’ve picked up several really good tips from this forum. Some very knowledgable folks on here.

Smuj

Foxjet · Jan 2, 2020

rwbaron said:
I used to do a lot of B&W in the film days but my only experience with digital B&W is with Canon APS C and X-Trans. My first 10+ years of digital were Canon and I never liked the results converting the Bayer raw files to B&W but X-Trans is a whole different story. I now shoot B&W quite a bit and first used Lr and then Tonality but have been more active recently with Silver Efex Pro 2.

Bob

Excellent results, Bob. I like working with Silver Efex Pro 2 as well.

Dusty-Lens · Jan 2, 2020

Tom Schum said:
Each of the pixels in the final image is the average of a group of four pixels in the original image. This basically eliminates the interpolation artifacts of CFA (Color Filter Array) sensors.

Doug MacMillan · Jan 2, 2020

TRIODEROB said:
any who - think about which of the sensor gets you closest to the real silver halide film look.

we have

1) Xtrans which is found in the Fuji cameras

This is my favorite for B&W. The Acros simulation with the filter choices and ability to tweak highlight and shadow settings allow me to easily approximate Plus-X, Tri-X, Agfa and of course Fuji.

TRIODEROB said:
2) Foveon which is found in Sigma cameras

I have now experience.

TRIODEROB said:
3) Bayer which is the normal type digital camera tech

With NIK Silver Efex Pro I can get really nice looking B&W. It's not as fun as shooting Fuji though.

TRIODEROB said:
4) dedicated Monochrom which is found in the Leica

I've tested with the Leica and really like the results, though it takes some work in NIK Silver Efex Pro to get good results. There's nothing like the experience of shooting with a Leica rangefinder, which I've been doing since 1968. Still, you have to filter when shooting, which is not as flexible is applying filters in post or being able to select the Fuji film sims filters.

TRIODEROB said:
perhaps one of you has experience in all 4 and can chime in - provided you also have experience in the silver halide process too …...

I developed my first roll of film in 1959 when I was 8. Since then I've shot thousands of rolls of B&W film, including the time I was a full time professional commercial and industrial photographer. I love B&W images. Here's what you can get with Fuji:

X-E3, Godox in Westcott Rapid Box camers right

Dusty-Lens · Jan 3, 2020

Would the correct workflow be to convert to b&w initially and then use /4 nearest neighbor downscaling or vice versa?

Will this work as expected if the vertical or horizontal pixel count isn't devidable by 2?

Thank you!

Foxjet · Jan 3, 2020

Doug MacMillan said:
TRIODEROB said:

any who - think about which of the sensor gets you closest to the real silver halide film look.

we have

1) Xtrans which is found in the Fuji cameras

Click to expand...

This is my favorite for B&W. The Acros simulation with the filter choices and ability to tweak highlight and shadow settings allow me to easily approximate Plus-X, Tri-X, Agfa and of course Fuji.

Doug MacMillan said:

2) Foveon which is found in Sigma cameras

Click to expand...

I have now experience.

Doug MacMillan said:

3) Bayer which is the normal type digital camera tech

Click to expand...

With NIK Silver Efex Pro I can get really nice looking B&W. It's not as fun as shooting Fuji though.

Doug MacMillan said:

4) dedicated Monochrom which is found in the Leica

Click to expand...

I've tested with the Leica and really like the results, though it takes some work in NIK Silver Efex Pro to get good results. There's nothing like the experience of shooting with a Leica rangefinder, which I've been doing since 1968. Still, you have to filter when shooting, which is not as flexible is applying filters in post or being able to select the Fuji film sims filters.

Doug MacMillan said:

perhaps one of you has experience in all 4 and can chime in - provided you also have experience in the silver halide process too …...

Click to expand...

I developed my first roll of film in 1959 when I was 8. Since then I've shot thousands of rolls of B&W film, including the time I was a full time professional commercial and industrial photographer. I love B&W images. Here's what you can get with Fuji:

X-E3, Godox in Westcott Rapid Box camers right

Doug, Great insights and photo, thanks for sharing.

--
Richard

Richard Bledsoe

Explore Richard Bledsoe’s 3,388 photos on Flickr!

www.flickr.com

Tom Schum · Jan 3, 2020

Dusty-Lens said:
Would the correct workflow be to convert to b&w initially and then use /4 nearest neighbor downscaling or vice versa?

Will this work as expected if the vertical or horizontal pixel count isn't devidable by 2?

Thank you!

I guess you would have to experiment with different workflows. I usually downsize then convert to monochrome and add grain (or not).

Nearest Neighbor might not work so well if vertical or horizontal pixel count is not dividable by 2, or 3, or 4, or 5 (or some integer). Try non-integer with Nearest Neighbor and make your own decision about this. I've seen some odd looking edge artifacts but these might not be an issue in your own work.

In these cases (non-integer) I have been using the Bilinear algorithm, then sharpening a little bit afterward. Best thing would be to experiment a while with the various algorithms and choose the one you like best overall.

For a long time I just went with Bicubic (the usual default) and I was fine with it.

Dusty-Lens · Jan 3, 2020

Thanks! I want to experiment with nearest neighbor /4 downscale to check how close it comes to non-bayer b&w.

My theory would be to go for b&w initially so that 4 pixels are already grayscale before being averaged to 1 as a reverse bayer filter. But I might be completely wrong as often happens.

Generally my default in PS is bicubic sharper or smoother depending on the downscaled image quality. I don't upscale photos. For video downscaling I use mitchell and for video upscaling I use ewa lanczos sharp.

TRIODEROB · Jan 3, 2020

for those interested here is a road I am thinking about traveling down with BW

"Monochrome2DNG"

( warning - involved and complex subject )

https://www.fastrawviewer.com/Monochrome2DNG

Dusty-Lens · Jan 4, 2020

Starting at hardware level and removing the CFA is the only way to make a bayer sensor closer to Foveon in B&W and note loose 4x of the image size. Thank you for sharing this.

I would have to go another way, since I can't afford myself to dedicate a camera just for b&w and remove the cfa at the moment.

baobob · Jan 5, 2020

My only experience is with my 2 bodies XT3 and XH1 I use mainly Silver Efex in PS to produce my BW photos

Sometimes my WF is Color then High key in Color Efex then Silver Efex then ACR filter in PS this WF allowing me to dose very precisely the mid tones as needed

below some samples

IMHO I think the sensor is not the determinant factor but the WF you use from your RAW file

bob

--
Good judgment comes from experience
Experience comes from bad judgment

Xtrans vs Foveon vs Bayer vs Monochrome for BW ..... ???

Community Leader

Senior Member

Forum Enthusiast

Senior Member

Community Leader

Forum Enthusiast

Leading Member

Attachments

Forum Pro

Well-known member

Senior Member

Senior Member

Senior Member

Senior Member

Senior Member

Forum Pro

Senior Member

Senior Member

Senior Member

Forum Pro

Similar threads

Keyboard shortcuts