Another one on resolution (Bayer vs. Foveon) - AND or OR?

Hi all,

I read through a good part of the other threads (some called them trolls, not quite sure what that really means, not being a native speaker of English), but since they became long and kind of dramatized, I'll start a new one here. There are a couple of thoughts that came to my mind. Here we go (for some of you again):

BAYER sensors:

Every one of the (bigger number of) pixel locations need to "ask" their neighbour what the colour should be, then the real resolution of the imager is less than that number.

FOVEON sensors:

Every one of the (smaller number of) pixel locations has three options to get the right colour without having to "ask" around, the real resolution is exactly that number and therefore the "ideal" (i.e. best per pixel) resolution, not diluted by guesswork as some have put it.

Now, being the owner of the DP2m and former owner of DP2s and 1x, I see the difference to the Bayer sensor's results quite clearly - and am amazed by it, not wasting any information IQ-wise, yet producing bigger files for a yet to be seen reason.

Now comes the AND or OR issue:

The issue that seems to come up time and again and that I haven't seen resolved yet is this: Sigma states a resolution of around 45 MP. Now, while it is true that mathematically, there is a total of 45 million spots for the photons to "settle", the sensor only has 15 million spots for them to go, meaning that within each of those 15 million, the light chooses the right depth in the 3 layers, making it rather an OR, whereas in the Bayer sensor it is an AND, yet with the guesswork which reduces the final IQ quite a bit.

My suggestion would therefore be to put it this way: The Bayer sensor just needs so much more pixels to match the "ideal" pixel count of the Foveon sensor - and that being only 15MP instead of 45MP - that it doubles or (according to others) triples the needed pixels in order to reach the same IQ.

Yet now comes along a new generation of cameras without that annoying AA-filter or at least with a lesser impact of the same, called the "Naked sensor" by Luminous Landscape (http://luminous-landscape.com/essays/the_naked_sensor.shtml).

I don't want to heat up an already heated discussion, but for a fair comparison of the DP#M sensor (including a great lens) with a Nikon D800e with 36MP Bayer sensor, shouldn't we admit that the Bayer does outresolve the Merrill by quite a bit in that it just has the incredible pixelcount PLUS the lack of the AA-filter that swallowed detail.

Finally I do have the question why filessizes on the DP series are so big. Maybe I am wrong with the assumption of the OR-argument, that would clear up the issue, but as I understand it, every pixel location only renderes one colour (hopefully the right one) within the three options, so why does it retain the amount of information for all three?

Anyway, don't let that spoil your year end fun activity. Thanks for reading all this and Happy New Year. Looking forward to your thoughts.

That is a nice thought but expect drama to return whenever you critique the bayer sensor on this board.

Read your bayer description again though.  It wasn't just the AA filter that swallowed detail - that was just part of it, it was an overlay on the CFA (color filter array) which is NOT removed.  It was the fundamental act of each real output pixel being derived from a number of other photosites nearby, not wholly data from that location.  That is true with or without the AA filter.  From reading your description again possibly you thought the AA filter was the same as the CFA?

As for the D800E outresolving by "quite a bit" - what have you seen that makes that apparent?  Nothing I have seen indicates that is so.   The key word is "detail" - when you up-sample a DP-2M image to match a D800E image size, you would find nearly equivilent details - that is anything you could discern in one image would be present in another (assuming you found a suitably sharp lens on the D800E to use in comparison).

Lastly of course, if you shoot a color resolution chart for some colors the D800E will not in fact outresolve the DP-2M.  The 36MP D800E has only 9 million photosites each for red and blue detection.  So to make a blanket claim that the D800E resolution "outresolves" the DP-2M is simply not accurate.

Over time it's come to be pretty accepted that on average, the Foveon sensor matches up with about the number of green detectors in a bayer sensor - 18 million for the D800e, which is why it's slightly sharper.  That's why I use the term 46MP (30 MP bayer equivalent).  But again that is only an average, and not true in all cases - fundamentally you are comparing two very different systems where in a Foveon image color does not affect detail captured and in the bayer system it does.  So any comparison leading to a single "this has more detail" conclusion is kind of doomed to failure unless the numbers are vastly different on one side or the other.

Because you need all three to put the actual color back together again, and it's better to put that information into a file pre-conversion in case you figure out how to do a better job of that later, or leave the work to a far more powerful desktop system with more advanced software than can be include in-camera.  Otherwise you may as well just shoot JPG if you trust the in-camera engine enough to do all the work.

Have a great new year also!  Happy shooting!

--
---> Kendall
http://www.flickr.com/photos/kigiphoto/
http://www.pbase.com/kgelner
http://www.pbase.com/sigmadslr/user_home

Can you please provide information on where Sigma states this?

--
Laurence
laurence at appledore-farm dot com
Sometimes, life is going to hit you with a brick. Don't lose faith.
Steve Jobs
http://www.pbase.com/lmatson
http://www.pbase.com/sigmadslr
http://www.howardmyerslaw.com

Darn, beat me to it.

Never seen that claim.

Richard

BTW Laurance, are we likely to see you at Focus again anytime?

--
Formerly posting as 'Blissfly'

Actually Sigma states that the effective resolution is 46 megapixels on their web site here:

http://www.sigmaphoto.com/shop/dp2-merrill-compact-digital-camera

Regards,

Chuck Lane

I'm sorry but I can't find a reference to resolution on that page, number of pixels yes, but not resolution.

Richard

--
Formerly posting as 'Blissfly'

(That you are a native speaker becomes irrelevant at times )

I thought it was a corruption of the word "trawl", like to look through threads and get what you can out of them without contributing much that is useful in return. However, Wikipedia seems to have some notions too, so...

http://en.wikipedia.org/wiki/Troll_(Internet)

and also this one...

http://www.urbandictionary.com/define.php?term=internet%20troll

atom14.

Just a couple points. First a point about "resolution." Even though the marketing departments of camera manufacturers use the term "resolution" and "megapixels" in the same breath as in: "This camera has 20 megapixels resolution," it is an improper use of the term.

The "only" thing which allows this to have meaning at all is that with CFA type cameras, changing the pixel count upward or downward results in a "similar" gain or loss in the percentage of optical resolution. This is only because these cameras have similar sensors and "most" have antialiasing filters as well. So when we compare two similar sensors with similar output pixel count the optical resolution is probably somewhat close.

The "proper" way to evaluate system (sensor plus lens) optical resolution is to photograph a standardized black and white and a standardized color resolution chart and determine the point at which no further detail can be seen. That point will be referenced by a number which tell the technician what the horizontal and vertical and perhaps diagonal resolution is.

When the first Sigma digital camera (the SD9) was measured, the "optical" black and white resolution for this so called (3.4 megapixel) camera was almost exactly equal to the measured optical resolution of the Canon EOSD60, a so called "six megapixel" camera. Since that time measurements using color resolution charts showed that the "color" optical resolution of the SD9 was better than the Canon D60. This is where the Foveon based Sigma 2X equivalency to CFA cameras in optical resolution was born. Since this time, the Foveon sensor has remained pretty well resolving to the Nyquist limits for both horizontal and vertical measurements. On the other hand, there is a degree of variance in CFA cameras depending on whether or not they have an AA filter and if so how strong. So "precise" equivalencies are not as simple to extract as they once were.

In addition to the differences described above, the Foveon sensor produces a naturally "sharper" image than similar "resolution" CFA based cameras because there are fewer pixels used in the roll-off at detail edge boundaries. This is true whether or not there is an antialiasing filter on the CFA camera. The CFA (Color Filter Array) itself contributes some of the additional edge roll-off pixels because of its inherent design and the interpolation necessary to guesstimate colors. Additionally, the micro-contrast is better with the Foveon sensors and therefore accutance is increased.

There are two reasons why the file size is appreciably larger in the Foveon capture. First, because there is more detail being captured. Yes, all three loci of RBG are used to create a single pixel value, but I understand that these values are stored in the file, not only the calculated pixel value. If you photograph a pure color with zero detail you will see very little differences in file sizes. As the amount of natural detail in the captured scene increases, so does the file size. Images with lots of micro detail (and thus lots of micro contrast) take more storage space and the detail increases. With the Bayer CFA capture, less actual detail (assuming similar true resolution) is captured and much of the chroma detail is actually "manufactured" during the interpolation process.

This is essentially what my experience has led me to understand about the differences - other's may have different explanations.

Best regards,

Lin

--
learntomakeslidshows.net

You're right of course. On Sigma's website they state 48 megapixels total and effectively 46 megapixels. I took the 'effective' part to mean resolution.

Regards,

Chuck Lane

Ask Graham and Paul

--
Laurence
laurence at appledore-farm dot com
Sometimes, life is going to hit you with a brick. Don't lose faith.
Steve Jobs
http://www.pbase.com/lmatson
http://www.pbase.com/sigmadslr
http://www.howardmyerslaw.com

Hi Chuck,

Still daft, I guess. I could not find that on your cite. Could you help me further please?

L

--
Laurence
laurence at appledore-farm dot com
Sometimes, life is going to hit you with a brick. Don't lose faith.
Steve Jobs
http://www.pbase.com/lmatson
http://www.pbase.com/sigmadslr
http://www.howardmyerslaw.com

Darn, beat me to it.

And so we play etherial ping-pong.

Your court.

--
Laurence
laurence at appledore-farm dot com
Sometimes, life is going to hit you with a brick. Don't lose faith.
Steve Jobs
http://www.pbase.com/lmatson
http://www.pbase.com/sigmadslr
http://www.howardmyerslaw.com

Thanks for figuring that out. Now you can help me straighten out the rest of the industry. Can you ride a donkey? I have a white horse and can joust with windmills very well.

--
Laurence
laurence at appledore-farm dot com
Sometimes, life is going to hit you with a brick. Don't lose faith.
Steve Jobs
http://www.pbase.com/lmatson
http://www.pbase.com/sigmadslr
http://www.howardmyerslaw.com

Ok, I'm on the donkey….now then, has anyone actually tested the DP2M for resolution?

Regards,

Chuck Lane

yes

--
Laurence
laurence at appledore-farm dot com
Sometimes, life is going to hit you with a brick. Don't lose faith.
Steve Jobs
http://www.pbase.com/lmatson
http://www.pbase.com/sigmadslr
http://www.howardmyerslaw.com

Originally it came from the fishing technique called "trolling" where you trail a bait slowly through the water and see what bites.  For example, someone might go on a sci-fi forum and say something like: "Didn't that Leonard Nimoy guy play Dr Spock in the original Star Wars?" and then sit back and watch the responses of the outraged fan-people: childish, but mostly harmless. Because it was called trolling, the people doing it were referred to as "Trolls", and the name sort of lost the connection with its root, and now it has become a general term for anything from mild fun-poking to serious abusive behavior.

Joe

Popular Photography tested the SD1 to find out it has resolution equivalent to a 15 megapixel standard sensor.

Maybe it is better to think about it this way:

The foveon color resolution is equal to the spatial resolution. Bayer color resolution is, for the green, only half the total number of pixels, and for blue and red only 1/4 the total number of pixels. So, bayer color resolution is quite poor by comparison, and the artifacts from "expanding" the color resolution to equal the number of pixels add even more junk into the fine details.

If this is true, why is it that foveon has not taken over the world? At this point the loss of resolution due to bayer or any CFA is not enough to bother most people. It certainly doesn't bother me (well, it only bothers me a little):





--
Tom Schum

Hi Tom,

Actually, DpReview measured the optical resolution of the SD1 and here was the quote:

"The crops above show that, freed from the blurring effect of an anti-aliasing filter, the SD1's Foveon X3 sensor resolves extremely highly in this test. It's capable of distinguishing the lines on our resolution chart right up to its Nyquist frequency, and continues to render line structure some way beyond. This 'false' detail is the flipside to the lack of an anti-aliasing filter, and while purists may object that it's not an accurate reflection of the scene, it does tend to help make real-world images look more detailed. Notably there's not even a hint of colour moiré, because the X3 sensor records full-colour information at every pixel location."

I don't know what Pop Photography did wrong, but their conclusion is flawed. No conventional 15 megapixel sensor resolves 3136 lines horizontal and vertical. I have a "very good" resolving 16 megapixel Nikon D7000 and my DP2 Merrill soundly thrashes it in detail in every respect. Nyquist on the Foveon is 3136 lines and the DP2 M has an "excellent" optic so resolves very cleanly to Nyquist. For example, DpReview tested the 21 megapixel Canon 1DS Mark III at 2700 lines absolute with extinction at 3300 lines. This means that only 2700 lines were really clean but that "some" detail was still visible at 3300 lines.  In addition to the very high resolution, the micro-contrast and edge roll-off of the Foveon make the rendered actual resolution detail and acutance superb.

Keep in mind that there is also a difference between resolving nine cleanly detailed lines at Nyquist and resolving nine blurry lines at the identical point on the scale. So even if a conventional sensor of say 24 megapixels resolves slightly more lines than the Foveon, if these lines are indistinct and riddled with color moire (which is the case more often than not) we might say that "technically" the hypothetical 24 megapixel sensor produces higher resolution, but that the detail visible with the hypothetical slightly lower resolution of the Foveon sensor is visually superior. Couple this with the much better enlargeability quotient of the Foveon and you have a tool which produces superior large prints. One can further refine this by saying that seeing "something" which is indistinct and fuzzy is not as good as seeing a detail feature which is clean enough to be identified for what it is. I like to use pine needles as an example. Often with my very good resolving Nikon D7000 distant pine needles become an indistinct "blur" of green mush while with my DP2 Merrill, they are distinctly identifiable for what they are.

Think back when German optics were supreme for resolution, but had low contrast. Japanese optics, conversely, had lower resolution but far superior contrast and were more highly revered by most photographers. There is more to image quality than strictly resolution.

Best regards,

Lin

--
learntomakeslidshows.net

Dear Kendall,
Thank you for your quick reply to my thoughts. Yes, you are right, I was not thinking of the CFA that has a draining effect on resolution. Thanks for pointing that out. I was indeed not differenciating between the AA filter and the CFA.

The idea of the D800E outresolving the Merrills "quite a bit" was not based on actual comparison of images, but on the conclusion that often I read that the actual Bayer equivalent of the Merrill's Foveon is about 24-30MP would give the D800E with its 36MP "quite a bit" of gain over it. But again, as you pointed out, only the number of green detectors on the D800E outcount the green detectors of the Merrill (15 million), not the other two colours.

So, my conclusion now is, both are hard to compare and differ from image to image (colour detail and distinguishable lines depend on a number of factors like contrast etc.)

All the best,

Mattanja

Dear all,

Thanks so much for clearing up some misunderstandings on my side. I did forget about the CFA when mentioning the AA filter only as the distinguishing factor between Foveon and Bayer. Also, I did learn, that pixel sites don't mean resolution, epecially with regards to the other system to which one wants to compare the values.

Also, the explanations for the bigger filesizes were helpful to me. Thanks for taking the time to answer these thoughts and questions of mine.

Best regards and wishes to everyone for 2013.

Mattanja