How do you Prove Foveon Image Quality is Higher than Bayer?

MyReality · Jul 25, 2017

I assumed that using the same lens was a given. The test would use Jpegs for standardization. Since we are testing different cameras, by implication we are testing different sensors. Dynamic range, color rendition noise, etc. are properties of the sensor and Jpeg processing. Each camera would be set to its same white balance and exposure. A studio scene or chart could be use to keep parameters to a minimum.

The point is to see if the print image quality from a Foveon is statistically better by a wide enough margin to make it worth purchasing above other cameras by image quality alone.

Pixel peeping is valid only as a point in technical discussions, not a practical matter. That argument never ends, since you can always find a better camera.

richard stone · Jul 25, 2017

JayDog said:
Richard I understand your point with actual numbers and absolute proof. It becomes more of a touchy feely type of thing.

That said, what I was really looking for was a response or reason people choose a tool for a specific job. It is shocking to me that in all the post there is not some photographer with some examples between cameras on why he chose a specific camera / lens. I understand if I am shooting flying birds I probably go with Canon/Nikon for the focus speed. I could probably back that up when shooting with a sigma camera at flying birds gets me 1/25 shots comes out clean vs 1/8 shots with faster focus Nikon gear.

Wedding photographers use certain gear and sub gear within that brand to get the shots they need.

View Cameras/Medium format all have their places.

Now what I was really expecting was a photographer to say I use to shoot canon and was testing out this new sigma gear and shot on top of a mountain and was blown away at the quality differences between the 2. Here are the photos and this happens every time ect.

From what I hear you saying the differences are so minor it is not really measurable or easy to spot. Please see my other posts of Jpegs 80d vs Quattro Simple Quick studio shots.. To be honest in jpeg mode I do not see a huge difference. In this post maybe somebody else has a lot better eyes.

To end this, I have seen large format prints, Truly crazy detail, Medium format very nice, excellent focus speed 35mm, very high dynamic range modes. I was expecting more from the forum group as a whole than well because it shoots better pictures. There is a lot of technical jargon in here but in the end it all comes down to obvious results.

Jay

Jay, I personally really like the images from the sdQ. I was very happy with the SD10...Until I saw what the Q sensor could deliver. The sdQ matches up well with the very best of the CFA cameras at low ISO. The H version is even better. Both work well for landscape images and city images, studio work and such, and to some extent for general use. I don't do weddings or sports, and if I did I would probably use a different camera. Probably: because with proper lighting I'm sure the Sigma would do fine.

What I see is a different "look" to the Sigma images, in many cases, but of course that can be changed, or lost, in PP. And for that matter, how much of a different look do we want? I mean, we are taking a picture of a real world scene, we sort of expect the picture to look real, like what we saw. But it is essentially impossible for the image to be exactly what we saw. All we can do is measure color accuracy and luminance detail. But check out the images on this site from Steve Chong and LWW.

Loading…

www.dpreview.com

Loading…

www.dpreview.com

The virtues of the Foveon sensor are also the lack of interpolation and the lack of a blurring filter. The images are inherently and uniformly sharper, meaning with more detail, in regard to all colors, compared to a mainly green filtered pixels of CFA. In theory the CFA cameras can do as well as the Q when the sensor has a great number of pixels.

The Merrill ("M") series of cameras is quite a bit different.

For that matter, it does not look to me like SPP is presently getting all it can from the Q sensor. If some of the recent PP threads are any indication, there is more to get out of those images, including a better DR.

As to proof, one is (or at least I am) reminded of the line from the treasure of the Sierra Madre, with regard to not needing stinkin' badges.

Richard

MyReality · Jul 26, 2017

Let me re-iterate. Obviously, each cameras parameters would be set as equally as possible. Noise reduction would be set the same or as similar as possible.

Using converters would invalidate the test. The point is the use the same camera and lens for the test. A converter may degrade the image differently for each camera.

I have no issues with the Foveon sensor tests against the 5DSR at the pixel level on a monitor as a subjective test with possibly 10 different answers,, but this does not prove anything.

The point is, can the differences at pixel level be seen in a same size practical print test. You can always make a print big enough to see the differences, but would you? Making a print from a standard DX sensor capture viewed at 100% makes a 5'x7' print.

danski0224 · Jul 27, 2017

richard stone said:
The sdQ matches up well with the very best of the CFA cameras at low ISO. The H version is even better.

Can you please expand on the "better" part of the Quattro H sensor?

I have been unable to find any differences between the standard Q or H versions, with the exception of size.

PrebenR · Jul 28, 2017

How do you prove reality is how we perceive it to be? ;-)

danski0224 · Jul 28, 2017

PrebenR said:
How do you prove reality is how we perceive it to be? ;-)

MyReality · Jul 28, 2017

I guess I need to clarify again, yes it depends on resolution, dpi and ppi and viewing distance of the print. I have known museum quality exhibitors who print at 150dpi for larger print. Billboard prints may be printed at much less than that.

What distance do you think a 5'x7' print would be view at? At that viewing distance they can be printed at much less than 150 dpi. Very good 20"x30" prints have been made from 6 meg files. Very good 24"x36" conventional prints have been made from Kodachrome 25 and Velvia 50.

Roland Karlsson · Jul 30, 2017

George_pix said:
If you are a mathematician you will become very irritated knowing that your camera attempts to solve a mathematical problem (demosaicing) that cannot be solved exactly by definition.

Or you find it fascinating!

George_pix said:
With a Merrill sensor you know that the above "problem" does not exist at all.

That is only so in a very simplified theory.

In practice, you have to solve the color problem and you have to fight with the noise problem. Actually, Merrill and also Quattro needs NLM non local means noise reduction, which absolutely not has one exact solution.

George_pix said:
With the "quattro" sensor the exact solution exists but the resolution reduces towards long wavelengths (red).

And this is even more simplified. We have had long discussions here when Quattro came how to do the interpolation. There is not one way.

All this said - I agree with Ted. The main difference is that Bayer has color aliasing.

Roland Karlsson · Jul 30, 2017

xpatUSA said:
That is the only reason for me - no color aliasing.

+1

Roland Karlsson · Jul 30, 2017

You have started a rather interesting thread. I have not read it all yet. I will just give two short experiences of mine.

I am mainly working with Bayer cameras. As good ones as I can afford. Currently my main camera is a Pentax K-1. Digital photography is fantastic for color photography. I really like it.

EXPERIENCE 1

The first time i saw very well made Foveon images, I said WOW. That was some clarity that I cannot really see with Bayer images. Remember, this was some years ago, so both Bayer and Foveon had fewer pixels. The difference is not that big any more. Still there though.

But, and here comes the catch. Looking closer at the super clear Foveon images they often look a bit strange. Details, that are a bit fuzzy on Bayer, looks a bit distorted on Foveon. This is mainly aliasing. On Merrill it is also caused by noise reduction IMHO.

So ... I went from WOW ... to ... Hmmmm.

Of course, making a huge print, maybe the WOW dominates.

EXPERIENCE 2

Bayer only sample one color at every pixel, so the samplings are out of phase and you get color aliasing. That is the major draw back of Bayer. You can improve this with more pixels, but really not fix it 100%. Otherwise, Bayer are very good at color accuracy as the color filters can be chosen with care.

Foveon solves the color aliasing problem. That can be very nice. It makes it possible to omit the AA filter without serious problems. Which makes sharper and clearer images (see above though).

But ... there are several new color problems.

One is blotches. Those blotches are created due to poor separation of color channels after conversion. The gray is not sure if it is green or magenta. Are they common? Nope, but they have a tendency to pop up when they are not welcome. Can you avoid them? Sure, I assume so. But I do not know how.

Another one is color fidelity. This is now much better. Quattro is very good. But (IMHO) older versions of Foveon has problem here. And it varies what problems from camera to camera. SD1 has a huge problem with red, that tends to be orange. Some older cameras have huge problems with green, that tends to be miscolored when photographing flowers. And I have even seen images of white beaches that are cyan. Bayer images tends to be cold or warm when they have problems with color, e.g. in mixed lighting. Which actually mostly look OK. But, Foveon images has a tendency to go away somewhere else in the color space, magenta, cyan, green, ... Which is much more annoying. As I said, I am not sure this problem exists with Quattro.

A third one is the need for NLM noise reduction. Without it, Merrill and Quattro would be useless. In our (Kalpanika) converter, you can turn off NLM. And, you will not like the result. Not at all. What is then NLM? It is a non local mean "smoothing". It works like normal smoothing by averaging, but you use weighted non local information instead of local. This leads to a non soft "smoothed" image, but at also means that the resulting image might be fake. I would not use Merrill or Quattro in scientific or medical applications. Because, the result might be a lie. Sometimes you can see this lie, when e.g. two green leaves are yellow because a nearby wall is yellow. Is this serious? Nope, not at all for artistic photography. You will not be able to notice it. I have only seen it twice.

PanoMax · Jul 30, 2017

Here are a few recent shots to show examples of image quality with the DP3 Merrill and it's Foveon sensor. The Merrill series produces different results than the Quattros.

No, it's not all things to all people... the camera is only as good as the operator and his understanding of it's limitations and how to control the many aspects of light.

Just like automobiles...don't expect to drive your Porsche up a logging road as you would with your Jeep.

db

TN Args · Jul 30, 2017

Roland Karlsson said:
You have started a rather interesting thread. I have not read it all yet. I will just give two short experiences of mine.

I am mainly working with Bayer cameras. As good ones as I can afford. Currently my main camera is a Pentax K-1. Digital photography is fantastic for color photography. I really like it.

EXPERIENCE 1

The first time i saw very well made Foveon images, I said WOW. That was some clarity that I cannot really see with Bayer images. Remember, this was some years ago, so both Bayer and Foveon had fewer pixels. The difference is not that big any more. Still there though.

But, and here comes the catch. Looking closer at the super clear Foveon images they often look a bit strange. Details, that are a bit fuzzy on Bayer, looks a bit distorted on Foveon. This is mainly aliasing. On Merrill it is also caused by noise reduction IMHO.

So ... I went from WOW ... to ... Hmmmm.

Of course, making a huge print, maybe the WOW dominates.

EXPERIENCE 2

Bayer only sample one color at every pixel, so the samplings are out of phase and you get color aliasing. That is the major draw back of Bayer. You can improve this with more pixels, but really not fix it 100%. Otherwise, Bayer are very good at color accuracy as the color filters can be chosen with care.

Foveon solves the color aliasing problem. That can be very nice. It makes it possible to omit the AA filter without serious problems. Which makes sharper and clearer images (see above though).

But ... there are several new color problems.

One is blotches. Those blotches are created due to poor separation of color channels after conversion. The gray is not sure if it is green or magenta. Are they common? Nope, but they have a tendency to pop up when they are not welcome. Can you avoid them? Sure, I assume so. But I do not know how.

Another one is color fidelity. This is now much better. Quattro is very good. But (IMHO) older versions of Foveon has problem here. And it varies what problems from camera to camera. SD1 has a huge problem with red, that tends to be orange. Some older cameras have huge problems with green, that tends to be miscolored when photographing flowers. And I have even seen images of white beaches that are cyan. Bayer images tends to be cold or warm when they have problems with color, e.g. in mixed lighting. Which actually mostly look OK. But, Foveon images has a tendency to go away somewhere else in the color space, magenta, cyan, green, ... Which is much more annoying. As I said, I am not sure this problem exists with Quattro.

A third one is the need for NLM noise reduction. Without it, Merrill and Quattro would be useless. In our (Kalpanika) converter, you can turn off NLM. And, you will not like the result. Not at all. What is then NLM? It is a non local mean "smoothing". It works like normal smoothing by averaging, but you use weighted non local information instead of local. This leads to a non soft "smoothed" image, but at also means that the resulting image might be fake. I would not use Merrill or Quattro in scientific or medical applications. Because, the result might be a lie. Sometimes you can see this lie, when e.g. two green leaves are yellow because a nearby wall is yellow. Is this serious? Nope, not at all for artistic photography. You will not be able to notice it. I have only seen it twice.

Nice post, good summary.

D Cox · Jul 31, 2017

The lichen shot is superb.

xpatUSA · Jul 31, 2017

TN Args said:
Roland Karlsson said:

You have started a rather interesting thread. I have not read it all yet. I will just give two short experiences of mine.

I am mainly working with Bayer cameras. As good ones as I can afford. Currently my main camera is a Pentax K-1. Digital photography is fantastic for color photography. I really like it.

EXPERIENCE 1

The first time i saw very well made Foveon images, I said WOW. That was some clarity that I cannot really see with Bayer images. Remember, this was some years ago, so both Bayer and Foveon had fewer pixels. The difference is not that big any more. Still there though.

But, and here comes the catch. Looking closer at the super clear Foveon images they often look a bit strange. Details, that are a bit fuzzy on Bayer, looks a bit distorted on Foveon. This is mainly aliasing. On Merrill it is also caused by noise reduction IMHO.

So ... I went from WOW ... to ... Hmmmm.

Of course, making a huge print, maybe the WOW dominates.

EXPERIENCE 2

Bayer only sample one color at every pixel, so the samplings are out of phase and you get color aliasing. That is the major draw back of Bayer. You can improve this with more pixels, but really not fix it 100%. Otherwise, Bayer are very good at color accuracy as the color filters can be chosen with care.

Foveon solves the color aliasing problem. That can be very nice. It makes it possible to omit the AA filter without serious problems. Which makes sharper and clearer images (see above though).

But ... there are several new color problems.

One is blotches. Those blotches are created due to poor separation of color channels after conversion. The gray is not sure if it is green or magenta. Are they common? Nope, but they have a tendency to pop up when they are not welcome. Can you avoid them? Sure, I assume so. But I do not know how.

Another one is color fidelity. This is now much better. Quattro is very good. But (IMHO) older versions of Foveon has problem here. And it varies what problems from camera to camera. SD1 has a huge problem with red, that tends to be orange. Some older cameras have huge problems with green, that tends to be miscolored when photographing flowers. And I have even seen images of white beaches that are cyan. Bayer images tends to be cold or warm when they have problems with color, e.g. in mixed lighting. Which actually mostly look OK. But, Foveon images has a tendency to go away somewhere else in the color space, magenta, cyan, green, ... Which is much more annoying. As I said, I am not sure this problem exists with Quattro.

A third one is the need for NLM noise reduction. Without it, Merrill and Quattro would be useless. In our (Kalpanika) converter, you can turn off NLM. And, you will not like the result. Not at all. What is then NLM? It is a non local mean "smoothing". It works like normal smoothing by averaging, but you use weighted non local information instead of local. This leads to a non soft "smoothed" image, but at also means that the resulting image might be fake. I would not use Merrill or Quattro in scientific or medical applications. Because, the result might be a lie. Sometimes you can see this lie, when e.g. two green leaves are yellow because a nearby wall is yellow. Is this serious? Nope, not at all for artistic photography. You will not be able to notice it. I have only seen it twice.

Click to expand...

Nice post, good summary.

+1

PrebenR · Jul 31, 2017

Roland Karlsson said:
This leads to a non soft "smoothed" image, but at also means that the resulting image might be fake. I would not use Merrill or Quattro in scientific or medical applications. Because, the result might be a lie.

Whilst with a camera where interpolation is needed it i not a lie?
For scientific purposes you'd use other technologies.

richard stone · Jul 31, 2017

PrebenR said:
Roland Karlsson said:

This leads to a non soft "smoothed" image, but at also means that the resulting image might be fake. I would not use Merrill or Quattro in scientific or medical applications. Because, the result might be a lie.

Click to expand...

Whilst with a camera where interpolation is needed it i not a lie?

For scientific purposes you'd use other technologies.

TN Args · Jul 31, 2017

PrebenR said:
Roland Karlsson said:

This leads to a non soft "smoothed" image, but at also means that the resulting image might be fake. I would not use Merrill or Quattro in scientific or medical applications. Because, the result might be a lie.

Click to expand...

Whilst with a camera where interpolation is needed it i not a lie?

Interpolation is the estimation of a value between two known values (known accurately enough to use). As long as we don't pretend to have increased resolution, there is no lie. It is the truth, to the resolution limit.

NLM is the opposite of interpolation. It says that our 'known' values are so corrupted that we don't know them well enough to use, and so we look all over the data space for clues to the nature and extent of corruption, and make a generalised correction which we then present as known values per location. Whilst crossing fingers. And toes.

PrebenR said:
For scientific purposes you'd use other technologies.

Digital cameras are used for science, extensively.

FDecker · Aug 1, 2017

Roland,

it is not the first time that you mention NLM as the only way to handle noise in Foveon pictures. I accept that you have a better insight into the Foveon RAW data than I have and I know that you have a good technical understanding in all these things. But is there any indication that SPP uses NLM other than the fact that the Kalpanika team went for it because it did work for your approach? I ask because it always sounds as if you know that for sure. For me it is just an educated guess. Do you have more information on the SPP code?

I always attributed the color bleeding to the Quattro architecture. I never saw it with the Merril sensor. NLM shouldn't turn green leaves yellow, simply because this is a large shift in color. NLM searches for similarities and due to that approach it shouldn't change the color by that much. Unless it is badly implemented.

And dear all,

especially in the light of this thread, I am surprised that nobody used the test shots provided by Imaging Resource. There are RAWs from the DP2Q and the SD1M. A very controlled environment and the framing is always the same so that the focal length of the lens is compensated. My only concern is that the pictures are always taken with early firmware on the camera and they don't re-do the tests with later firmware versions (which of course would be a huge workload to do).

I played around with the files yesterday converting them with SPP 6.5.3. At low ISO, the Q wins in my personal point of view. At high ISO (1600), the Q is much worse than the M. Especially the red fabric is just some red "nutella" with the Q. And overall at each sharp color transition, one can see artifacts which look like a combination of color bleeding and noise. There might have been firmware improvement since then. But, honestly, I was kind of shocked. Wasn't the Q designed to mitigate the problems with noise in the lower layers?

Now, last comment, what does it mean when the OP says IQ? There are some properties one can measure as MTF, S/N, DR, color fidelity, etc... But how these properties turn into a better IQ is not unambiguous and depends on the individual preferences of the observer. While it might be possible to prove that single properties are better with a given camera, we typically never get all of them optimal in one camera. So, there is always a choice to take by the designer and if the choice fits one's preferences one may say that the IQ is higher. But it may be lower for someone else. There will be no prove.

TN Args · Aug 1, 2017

FDecker said:
....i I am surprised that nobody used the test shots provided by Imaging Resource. There are RAWs from the DP2Q and the SD1M. A very controlled environment and the framing is always the same so that the focal length of the lens is compensated....

Good idea but use sdQ since the dp2Q raw is very early Quattro and updates to both the raw files and SPP mean it is not relevant today. Cheers

FDecker · Aug 1, 2017

TN Args said:
FDecker said:

....i I am surprised that nobody used the test shots provided by Imaging Resource. There are RAWs from the DP2Q and the SD1M. A very controlled environment and the framing is always the same so that the focal length of the lens is compensated....

Click to expand...

Good idea but use sdQ since the dp2Q raw is very early Quattro and updates to both the raw files and SPP mean it is not relevant today. Cheers

I processed the RAW with SPP 6.5.3. So, this is OK, then. I agree to the firmware point as I already wrote in my post. Unfortunately, there are no pictures of their test scene for the sd Quattro (H or non-H).

How do you Prove Foveon Image Quality is Higher than Bayer?

Veteran Member

Veteran Member

Veteran Member

Senior Member

Veteran Member

Senior Member

Veteran Member

Forum Pro

Forum Pro

Forum Pro

Leading Member

Attachments

Forum Pro

Forum Pro

Forum Pro

Veteran Member

Veteran Member

Forum Pro

Senior Member

Forum Pro

Senior Member

Similar threads

Keyboard shortcuts