Posts: 6,715
Threads: 236
Joined: Apr 2010
Reputation:
22
[quote name='boren' timestamp='1285952001' post='3434']
I'd be very interested to see real-life review that show that aliasing is indeed an issue with Foveon. I'm sure the effect can be demonstrated if one deliberately sets up the right conditions, subject, lighting and so on, but would it be visible in real life samples (as those in the end of the review at dpreview.com)?
Also, can you comment on the other points: "At what print size and viewing distance do you expect to be able to see it? What percentage of viewers do you feel would notice it?"
[/quote]
I am not against foveon sensors, Sigma cameras or AA-filterless cameras at all. I am merely explaining that the extra "sharpness" is an effect of the lack of AA-filter, not so much of it not being a non-bayer CFA sensor.
Because of the low resolution of the 3 and 4mp Sigmas, it would be apparent in relatively small print sizes, depending on the subject (straight lines in buildings/cars, silhouettes, hair etc.).
With the new 15mp camera, for the aliasing to show up you do need to go for bigger prints, obviously. Exactly the same as oversharpening...
But, the perceived extra sharpness will only show up with bigger prints too...
There just are people who like shooting without AA-filter. I am not one of them. And the foveon sensor has a bigger problem than aliasing: due to the nature of the sensor it will be more noisy at higher ISO settings.
genotypewriter
Unregistered
[quote name='Brightcolours' timestamp='1285930459' post='3420']
It is not just a claim. Sharpness does not come from the colour information, but from luminance information. The luminance information is based on the information per pixel, mostly.
So, yes, a Bayer sensor without AA-filter gets about the same resolution as a foveon sensor without AA-filter.
[/quote]
You mean when you compare an "X" megapixel Bayer CFA to a X/3 megapixel Foveon, right?
[quote name='Brightcolours' timestamp='1285952639' post='3435']
I am not against foveon sensors, Sigma cameras or AA-filterless cameras at all. I am merely explaining that the extra "sharpness" is an effect of the lack of AA-filter, not so much of it not being a non-bayer CFA sensor.
[/quote]
Sorry but that's absolutely not true and there are heaps of evidence out there. To start with, have you used an AA-filterless Bayer camera?
[quote name='Brightcolours' timestamp='1285952639' post='3435']
And the foveon sensor has a bigger problem than aliasing: due to the nature of the sensor it will be more noisy at higher ISO settings.
[/quote]
That maybe so but most people don't know how to compare resolution between a Foveon and a Bayer image with the resolution equalised. Otherwise it's just like saying the 5DII has more noise than a D3... of course because one has more detail than the other but when you equalise the resolution, that is not the case anymore.
Posts: 6,715
Threads: 236
Joined: Apr 2010
Reputation:
22
[quote name='genotypewriter' timestamp='1286027089' post='3452']
You mean when you compare an "X" megapixel Bayer CFA to a X/3 megapixel Foveon, right?
[/quote]
No I do not mean that.
[quote name='genotypewriter' timestamp='1286027089' post='3452']
Sorry but that's absolutely not true and there are heaps of evidence out there. To start with, have you used an AA-filterless Bayer camera?
[/quote]
It is very true. Of course, you do need to use a good RAW converter in the first place. Care to share the heaps of evidence?
[quote name='genotypewriter' timestamp='1286027089' post='3452']
That maybe so but most people don't know how to compare resolution between a Foveon and a Bayer image with the resolution equalised. Otherwise it's just like saying the 5DII has more noise than a D3... of course because one has more detail than the other but when you equalise the resolution, that is not the case anymore.
[/quote]
[quote name='Brightcolours' timestamp='1285940746' post='3425']
All "sites" deliver luminance information. Only interpolation of the data can give colour. It depends on the RAW converter just how much luminance information is being extracted, some are better at that than others. And of course, interpolation between the sensel values play part in it.
But it is wrong to think that because of interpolation luminance information (or data) is half of that of foveon sensors, that is simply untrue and just shows a lack of insight in the matter.
Which again still misses the major point: the sharpness of AA-filterless sensors is mostly FAKE detail. That you can measure fake detail as normal detail and think you get a lot higher resolution is of course true. Does not make the validity of the detail higher, though.
[/quote]
So you're saying that an array of four (GRGB) sites, sensitive to a specific wavelength and simply discarding the remaining data, will resolve exactly the same detail, as an array of four sites that collect 100% of light, right? Even if a high-contrast transition (let's say, between a red object and a blue sky) falls on a green pixel and the "edge" gets interpolated? Sorry, but it's completely against my understanding of how the interpolation works and unless you could provide some link to back you up, I find your claims completely unfounded.
Posts: 6,715
Threads: 236
Joined: Apr 2010
Reputation:
22
[quote name='Lomskij' timestamp='1286118565' post='3461']
So you're saying that an array of four (GRGB) sites, sensitive to a specific wavelength and simply discarding the remaining data, will resolve exactly the same detail, as an array of four sites that collect 100% of light, right? Even if a high-contrast transition (let's say, between a red object and a blue sky) falls on a green pixel and the "edge" gets interpolated? Sorry, but it's completely against my understanding of how the interpolation works and unless you could provide some link to back you up, I find your claims completely unfounded.
[/quote]
Of course you can think of very specific examples where you get a lower detail resolution. We are not talking about that, now are we? No, we are talking about detail in general. The point is (was): The main "advantage" of the foveon sensors (other than theoretical more precise colour infor per pixel) is that Sigma chooses not to use AA-filters, which makes them seemingly more sharp (but that sharpness is mostly fake detail, aliasing).
Don't make a caricature of things.
[quote name='Brightcolours' timestamp='1286127597' post='3463']
Of course you can think of very specific examples where you get a lower detail resolution. We are not talking about that, now are we? No, we are talking about detail in general. The point is (was): The main "advantage" of the foveon sensors (other than theoretical more precise colour infor per pixel) is that Sigma chooses not to use AA-filters, which makes them seemingly more sharp (but that sharpness is mostly fake detail, aliasing).
Don't make a caricature of things.
[/quote]
Right, right. The problem I see here is that you rather conveniently put an '=' sign between luminance data and image sharpness. To my understanding, it's 100% true in only one possible scenario - when someone shoots a pure B/W scene. Because as soon as you introduce any details based on colour transition (i.e. wavelength change instead of brightness change), demosaicing guesswork kicks in and resolution will degrade. I guess pure B/W scenes (the best case scenario for Bayer) occur naturally about as frequent as my "caricature" worst-case red-and-blue scenario, so if we assume that typical scene is somewhere in the middle, Bayer normally will have worse sharpness than Foveon. With AA filter or without.
P.s. I think we're back to square one here - I believe at the beginning of the conversation I agreed that Bayer would be equally sharp if it's B/W.
Posts: 6,715
Threads: 236
Joined: Apr 2010
Reputation:
22
[quote name='Lomskij' timestamp='1286179445' post='3471']
Right, right. The problem I see here is that you rather conveniently put an '=' sign between luminance data and image sharpness. To my understanding, it's 100% true in only one possible scenario - when someone shoots a pure B/W scene. Because as soon as you introduce any details based on colour transition (i.e. wavelength change instead of brightness change), demosaicing guesswork kicks in and resolution will degrade. I guess pure B/W scenes (the best case scenario for Bayer) occur naturally about as frequent as my "caricature" worst-case red-and-blue scenario, so if we assume that typical scene is somewhere in the middle, Bayer normally will have worse sharpness than Foveon. With AA filter or without.
P.s. I think we're back to square one here - I believe at the beginning of the conversation I agreed that Bayer would be equally sharp if it's B/W.
[/quote]
Your example (pure red and pure blue next to eachother) was silly. That was my point. It is not true that only black and white images show more defined pixels, and colour images do not.
Here the Nikon D200 as example:
http://www.maxmax.com/nikon_d200hr.htm
[quote name='Brightcolours' timestamp='1286193725' post='3475']
Your example (pure red and pure blue next to eachother) was silly. That was my point. It is not true that only black and white images show more defined pixels, and colour images do not.
Here the Nikon D200 as example:
http://www.maxmax.com/nikon_d200hr.htm
[/quote]
Quite interesting link. However unless we're looking into completely different pictures, I can see lots of examples that show nice and sharp black-to-white transitions and really, really bad rendering of colour edges with lots of ugly interpolation artefacts. That red speedometer needle is the best example imho - tack sharp black numbers and ragged red needle. As I can't see similar issues with Sigma's pictures, I'd say this link illustrates my point perfectly.
Posts: 6,715
Threads: 236
Joined: Apr 2010
Reputation:
22
[quote name='Lomskij' timestamp='1286198708' post='3477']
Quite interesting link. However unless we're looking into completely different pictures, I can see lots of examples that show nice and sharp black-to-white transitions and really, really bad rendering of colour edges with lots of ugly interpolation artefacts. That red speedometer needle is the best example imho - tack sharp black numbers and ragged red needle. As I can't see similar issues with Sigma's pictures, I'd say this link illustrates my point perfectly.
[/quote]
We are not looking at different things. I seem to be looking at what we can see, you seem to not be doing that.
We see much more detail all over, and yes, we see some colour artifacts from a bad RAW converter. We were not discussing the colour, though, we were discussing the extra "detail", which is very apparent (it is in luminance, not in colour space... both get extracted from the RAW data in different ways).
genotypewriter
Unregistered
[quote name='Brightcolours' timestamp='1286032045' post='3454']
It is very true. Of course, you do need to use a good RAW converter in the first place. Care to share the heaps of evidence?
[/quote]
I did, in my previous post. Let's look at this another way... keeping the fact that some AA-filters are overly strong, are you saying that Bayer CFA manufacturers don't want people to get more detail out of their sensors?
[quote name='Brightcolours' timestamp='1286193725' post='3475']
Here the Nikon D200 as example:
http://www.maxmax.com/nikon_d200hr.htm
[/quote]
The technical correctness of that test is questionable. Here are the reasons:
1. They're comparing two different cameras here. One is modified with a new clear glass filter which most probably has a different thickness to what the standard camera has. How accurately did they re-shim their focusing screen? And did they check the shimming of the stock camera's focusing screen to ensure that both produce the same level of accuracy (after all, manufacturing tolerances are normal).
2. These cameras don't have LV, so super-precise focusing necessary for focusing distances and magnifications like that is impossible... plain and simple. With or without eye-based focus bracketing, which is not accurate.
3. The AA-filterless camera here is what they refer to as a "hot rod" ("HR" postfix in the names, etc.) which is a full spectrum camera. Anyone who has experience with full-spectrum and IR imaging (/me raises both hands) knows that this introduces too many extra variables here.
3a. For one thing, most lenses focus UV, visible and IR light in to slightly different places (like multi-spectral CA) and only visible light can be seen at the time of focusing (obviously) but the final image contains UV and IR too.
3b. Then does the ICF they're using on the modified camera have the same spectral transmission curve as the ICF in the unmodified camera? Short answer: No. So what are we really seeing here (refer to 3a)? And on a side note, how did they manage to match the colours between the two so accurately when these image-capturing differences exist.
4. Not all AA filters are made equally. This is talking about a Nikon and a D200 of all things... technology has come a long way since then. Also according to what science is this observation going to be true for any other camera model? AA filters are different on every sensor. Since the D200 was available, for example, Canon has had the 5D (12MP FF), 30D (8MP), 40D (10MP), 450D (12MP), 50D (15MP), 60D (18MP), 1D2 (8MP APS-H), 1D3 (10MP APS-H), 1D4 (16MP APS-H), 5D2 (21MP FF), 1Ds3 (21MP FF/AA filter diff. from 5D2) all of which have their own unique AA filter. That's 11 different AA filters. Are you suggesting that all 11 of these were made like the D200's AA filter?
For me the biggest weakness in this test is what I mentioned in #2 above. Don't forget that these maxmax guys are also trying to sell their service...
GTW
|
