Canon's Rumored 75mp Camera To Have a Non Bayer Multilevel Sensor?

[Drizzt321]

I know it will never happen, but it would be so cool to see digital bodies become more film like. I'm talking about interchangeable sensors. Obviously, sensors work in concert with other integrated circuits so it would have to be an interchangeable sensor/processor board with a common bus to work with various bodies.

Actually you can buy just that, even trade in outdated sensors at a vastly more attractive rate then their retail cost.
Thats how things work in 645, even rather cross brand friendly. But try to explain that to the average store chain customer who is already uneasy with the interchangeable lens paradigm.

Wasn't there some announcement a while ago that had a lens+sensor module that was interchangeable? The way they were talking you could put in a FF with a wide/super-wide or with a very fast lens for landscape or low light, and then a moderate focal length with a very small sensor for super-telephoto at much less than a FF + 400/500/600mm lens. Or was it just an idea announcement and never got any attraction or went anywhere?

 

[dtaylor]

What they actually measure is aliasing:

The pattern is resolved. It's not perfect and there is aliasing, but it's there.

 

[Pi]

What they actually measure is aliasing:

The pattern is resolved. It's not perfect and there is aliasing, but it's there.

What if there is no pattern but fine detail, like in landscape photos?

Look at the resolution chart. It is an aliasing/moire fest, you can get dizzy just by looking at it.

EDIT: Here: http://www.sigma-sd.com/SD1/sample-photo/img/SDIM8118AA.jpg is an extreme (IMO) case of aliasing, taken directly from the Sigma webpage. Look the highlights in the hair, on (our) right of the chin and the neck. Is that resolution? How many "Bayer mp" of it?

 

[Loswr]

I am afraid that you are the one oversimplifying it.

Perhaps not eventually. Certainly for now, it's speculation and WAGs. But should Canon actually come out with a 75 MP (25 MP) Foveon-type sensor, we could measure LP/mm and/or SQF directly (i.e., with Imatest), and compare that to Bayer sensors with the exact same lens. Or maybe not - I'm assuming data from existing FF sensors can be plotted and extrapolated to simulate higher MP sensors that don't exist, but I haven't actually tried it.

Perhaps you can then tell me which is sharper: the 5D or the 60D? I have seen a comparison on dpreview which can support either conclusion, depending what you are looking at (and it is not AF or DOF problem).

That's why I stated 'existing FF sensors' - I'm takling about comparing resolution values from the same size sensor.

 

[dgatwood]

... a 75 MP (25 MP) Foveon-type sensor ...

I find that wording curious. I believe the original rumor said 75 million photosites, not 75 MP. A photosite is, by definition, a pit on the surface of the chip, not a subpixel. Thus, you would not divide by three, but rather optionally multiply by three to get an arguably disingenuous megapixel count.

 

[Loswr]

... a 75 MP (25 MP) Foveon-type sensor ...

I find that wording curious. I believe the original rumor said 75 million photosites, not 75 MP. A photosite is, by definition, a pit on the surface of the chip, not a subpixel. Thus, you would not divide by three, but rather optionally multiply by three to get an arguably disingenuous megapixel count.

True. I was merely echoing the title of the topic, "Canon's Rumored 75mp Camera To Have a Non Bayer Multilevel Sensor?"

 

[Pi]

I am afraid that you are the one oversimplifying it.

Perhaps not eventually. Certainly for now, it's speculation and WAGs. But should Canon actually come out with a 75 MP (25 MP) Foveon-type sensor, we could measure LP/mm and/or SQF directly (i.e., with Imatest), and compare that to Bayer sensors with the exact same lens. Or maybe not - I'm assuming data from existing FF sensors can be plotted and extrapolated to simulate higher MP sensors that don't exist, but I haven't actually tried it.

Perhaps you can then tell me which is sharper: the 5D or the 60D? I have seen a comparison on dpreview which can support either conclusion, depending what you are looking at (and it is not AF or DOF problem).

That's why I stated 'existing FF sensors' - I'm takling about comparing resolution values from the same size sensor.

I tried that once but I do not keep the result. If I remember well, each decent lens can at least double or triple its resolution with a super high mp sensor, in the (old) DXO units, for example. It was not just interpolation, it was based on a approximate formula how lens and sensor resolution act together. When I have more time, I will do it more carefully, with more data, if I can dig up the old DXO MTF tables.

Visual examples have been posted here, I believe. Pentax Q can do wonders with an EF lens. I do not have the time now to compute its pixel density, but it is certainly more than 75mp on FF.

 

[jrista]

I am afraid that you are the one oversimplifying it.

Perhaps not eventually. Certainly for now, it's speculation and WAGs. But should Canon actually come out with a 75 MP (25 MP) Foveon-type sensor, we could measure LP/mm and/or SQF directly (i.e., with Imatest), and compare that to Bayer sensors with the exact same lens. Or maybe not - I'm assuming data from existing FF sensors can be plotted and extrapolated to simulate higher MP sensors that don't exist, but I haven't actually tried it.

Perhaps you can then tell me which is sharper: the 5D or the 60D? I have seen a comparison on dpreview which can support either conclusion, depending what you are looking at (and it is not AF or DOF problem).

That's why I stated 'existing FF sensors' - I'm takling about comparing resolution values from the same size sensor.

I tried that once but I do not keep the result. If I remember well, each decent lens can at least double or triple its resolution with a super high mp sensor, in the (old) DXO units, for example. It was not just interpolation, it was based on a approximate formula how lens and sensor resolution act together. When I have more time, I will do it more carefully, with more data, if I can dig up the old DXO MTF tables.

Visual examples have been posted here, I believe. Pentax Q can do wonders with an EF lens. I do not have the time now to compute its pixel density, but it is certainly more than 75mp on FF.

Looks like the Pentax Q has 1.54µm pixels in a 1/2.3" sensor (6.17mm x 4.55mm). A 36x24mm sensor with the same pixel pitch would have around 23380x15600 pixels, which comes out to 365mp.

 

[CarlTN]

Does Sigma not hold the patent on the technique?

That would depend on what exactly was patented. The more you go into detail the easier its to do the same thing just different enough, the other way around you have to deal with prior art or significance.
Patents protect someones commercial interests, in research you still can use everything in there - one reason some inventions aren't patented: you gamble on the actual secret staying a secret instead of telling them exactly what they aren't allowed to do...for a while.

Interesting point.

 

[CarlTN]

What they actually measure is aliasing:

The pattern is resolved. It's not perfect and there is aliasing, but it's there.

What if there is no pattern but fine detail, like in landscape photos?

Look at the resolution chart. It is an aliasing/moire fest, you can get dizzy just by looking at it.

EDIT: Here: http://www.sigma-sd.com/SD1/sample-photo/img/SDIM8118AA.jpg is an extreme (IMO) case of aliasing, taken directly from the Sigma webpage. Look the highlights in the hair, on (our) right of the chin and the neck. Is that resolution? How many "Bayer mp" of it?

I'm not seeing where the aliasing in the hair is. In the part that goes out of focus behind the focal plane? I guess there is some. It's a remarkably sharp picture, though. She should have trimmed her nose hairs...!

 

[AlanF]

[/quote]

Looks like the Pentax Q has 1.54µm pixels in a 1/2.3" sensor (6.17mm x 4.55mm). A 36x24mm sensor with the same pixel pitch would have around 23380x15600 pixels, which comes out to 365mp.
[/quote]

The Powershot SX50 has basically the same size sensor (6.16mm x 4.62mm) with 1.54 µM pixels. Stitch 5x5 together to get an FF with about a 400MB file size for each RAW image.

 

[Pi]

I'm not seeing where the aliasing in the hair is. In the part that goes out of focus behind the focal plane? I guess there is some. It's a remarkably sharp picture, though. She should have trimmed her nose hairs...!

What you consider a remarkably sharp (hair) is a classical example of aliasing, with blown pixels, etc.

 

[jrista]

I'm not seeing where the aliasing in the hair is. In the part that goes out of focus behind the focal plane? I guess there is some. It's a remarkably sharp picture, though. She should have trimmed her nose hairs...!

What you consider a remarkably sharp (hair) is a classical example of aliasing, with blown pixels, etc.

Hmm, I see aliasing if I view the image at the scaled down size in my browser. When I click it to view at 100% size, that horrible aliasing disappears. The photo IS remarkably sharp, and there is only noticeable aliasing in a few areas where the highlights on each strand blow out. I would also point out that its "clean" aliasing...not the kind of color moire you get with a bayer sensor.

If you are trying to use that photo as an example of how bad layered sensor designs are...then at their worst, they are pretty darn good! I'm now completely sold on the approach. I can't wait to see a 3-layer sensor from Canon!

 

[Pi]

What you consider a remarkably sharp (hair) is a classical example of aliasing, with blown pixels, etc.

Hmm, I see aliasing if I view the image at the scaled down size in my browser. When I click it to view at 100% size, that horrible aliasing disappears. The photo IS remarkably sharp, and there is only noticeable aliasing in a few areas where the highlights on each strand blow out. I would also point out that its "clean" aliasing...not the kind of color moire you get with a bayer sensor.

Here is what I am talking about:

As if she suffers from dandruff. Looks horrible to me.

If you are trying to use that photo as an example of how bad layered sensor designs are...then at their worst, they are pretty darn good! I'm now completely sold on the approach. I can't wait to see a 3-layer sensor from Canon!

My remark has nothing to do with layered sensors. They need AA filters (at the current pixel densities) as well, just weaker ones.

 

[jrista]

What you consider a remarkably sharp (hair) is a classical example of aliasing, with blown pixels, etc.

Hmm, I see aliasing if I view the image at the scaled down size in my browser. When I click it to view at 100% size, that horrible aliasing disappears. The photo IS remarkably sharp, and there is only noticeable aliasing in a few areas where the highlights on each strand blow out. I would also point out that its "clean" aliasing...not the kind of color moire you get with a bayer sensor.

Here is what I am talking about:

As if she suffers from dandruff. Looks horrible to me.

If you are trying to use that photo as an example of how bad layered sensor designs are...then at their worst, they are pretty darn good! I'm now completely sold on the approach. I can't wait to see a 3-layer sensor from Canon!

My remark has nothing to do with layered sensors. They need AA filters (at the current pixel densities) as well, just weaker ones.

Hmm, I guess I wouldn't call what I see in that crop moire or aliasing. The edges of the hairs outside of the specular highlights are quite smooth...you can BARELY see some jagged edges in a couple of them, but for the most part, they appear to be fairly crisp, clean, and smooth. I think the issue there is just the nature of the lighting and the nature of her hair. I've seen similar issues with the D800 (non-E version WITH the AA filter). I would be willing to bet that some softening of the light would reduce the harshness of those highlights, and eliminate what you are calling aliasing.

 

[Pi]

Hmm, I guess I wouldn't call what I see in that crop moire or aliasing. The edges of the hairs outside of the specular highlights are quite smooth...you can BARELY see some jagged edges in a couple of them, but for the most part, they appear to be fairly crisp, clean, and smooth. I think the issue there is just the nature of the lighting and the nature of her hair. I've seen similar issues with the D800 (non-E version WITH the AA filter). I would be willing to bet that some softening of the light would reduce the harshness of those highlights, and eliminate what you are calling aliasing.

It is not "what I call aliasing", it is aliasing. The latter is not just moire of jagged lines. Extreme contrast between pixels is an artifact, always. Or maybe the model just needs to take a shower, I do not know. In any case, what I see looks awful.

EDIT: here is a computer generated aliased image, with no regular patterns. Bad hair or ... aliasing?

Source: http://www.howtogeek.com/73704/what-is-anti-aliasing-and-how-does-it-affect-my-photos-and-images/

 

[jrista]

Hmm, I guess I wouldn't call what I see in that crop moire or aliasing. The edges of the hairs outside of the specular highlights are quite smooth...you can BARELY see some jagged edges in a couple of them, but for the most part, they appear to be fairly crisp, clean, and smooth. I think the issue there is just the nature of the lighting and the nature of her hair. I've seen similar issues with the D800 (non-E version WITH the AA filter). I would be willing to bet that some softening of the light would reduce the harshness of those highlights, and eliminate what you are calling aliasing.

It is not "what I call aliasing", it is aliasing. The latter is not just moire of jagged lines. Extreme contrast between pixels is an artifact, always. Or maybe the model just needs to take a shower, I do not know. In any case, what I see looks awful.

EDIT: here is a computer generated aliased image, with no regular patterns. Bad hair or ... aliasing?

Source: http://www.howtogeek.com/73704/what-is-anti-aliasing-and-how-does-it-affect-my-photos-and-images/

Well, if we are going to get technical, technically speaking every pixel in an image is an alias of the real image the sensor resolved. Artifact or not, to me, the girls hair just looks crisp and sharp, maybe a bit oily or loaded with some kind of hair product, with some harsh specular highlights.

Comparing the results of a layered sensor without an AA filter to a bayer sensor with an AA filter, I think there is a definite sharpness benefit with the layered/no-AA approach, mostly offered by the fact that 100% of the image is being fully sampled at every photosite, vs. bayer which is sparsely sampled. If one finds it too sharp, there are a variety of ways that could be addressed....a mesh filter on the lens, post-process desharpening, etc.

Maybe this is just me, but I could never complain about too much sharpness. I am far more likely to complain about to little sharpness, as you really can't fabricate additional real detail in post.

 

[Pi]

Well, if we are going to get technical, technically speaking every pixel in an image is an alias of the real image the sensor resolved.

Google "sampling theory".

Comparing the results of a layered sensor without an AA filter to a bayer sensor with an AA filter, I think there is a definite sharpness benefit with the layered/no-AA approach, mostly offered by the fact that 100% of the image is being fully sampled at every photosite, vs. bayer which is sparsely sampled.

As I said, this is not a layered vs. Bayer sensor discussion.

If one finds it too sharp, there are a variety of ways that could be addressed....a mesh filter on the lens, post-process desharpening, etc.

Why not an AA filter instead of a mesh filter? PP cannot remove effects of improper sampling.

Maybe this is just me, but I could never complain about too much sharpness. I am far more likely to complain about to little sharpness, as you really can't fabricate additional real detail in post.

It is not detail. So you actually prefer this:

to that:

 

[jrista]

Well, I think your last comparison there is a bit unrealistic. You are comparing something on the order of 100x100 pixels to a photo with millions of pixels. I don't see the same thing (that rogue aliasing) in the original photo...its crisp and sharp, but not hideously aliased like your mock example. I've seen far worse examples of real-world aliasing and moire.

I'll see if I can find some examples of the D800 portraits I saw last year that were EXTREMELY sharp like that...and it seemed to be exactly what the photographer wanted. Sure, maybe its "aliased", but if that aliasing falls within the threshold of the photographers tolerance, I am not sure I would call it "incorrectly sampled".

I think at this point we are just debating personal preference, not any kind of real-world IQ-debilitating issue. Differing degrees of how sharp is sharp enough.

(CAVEAT: I just realized that I am looking at that photo on a MacBook 15" Retina. The pixel density may be changing how I perceive the photo. With this pixel density, it looks rather phenomenal...I'll admit it may look worse on a lower density screen. This screen is much closer to what a print would look like, and I would be quite happy with this kind of sharpness in a print.)

 

[Pi]

Well, I think your last comparison there is a bit unrealistic. You are comparing something on the order of 100x100 pixels to a photo with millions of pixels. I don't see the same thing (that rogue aliasing) in the original photo...its crisp and sharp, but not hideously aliased like your mock example. I've seen far worse examples of real-world aliasing and moire.

Actually, the Sigma photo is not less aliased, both compared at 100%. Think of this example as a crop of a wide angle shot with a lot of people there. Also, the Sigma photo looks unnatural even downsized a lot. Anyway, which image do you prefer?

I'll see if I can find some examples of the D800 portraits I saw last year that were EXTREMELY sharp like that...and it seemed to be exactly what the photographer wanted. Sure, maybe its "aliased", but if that aliasing falls within the threshold of the photographers tolerance, I am not sure I would call it "incorrectly sampled".

I think at this point we are just debating personal preference, not any kind of real-world IQ-debilitating issue. Differing degrees of how sharp is sharp enough.

We are talking about sampling theory. Is is a personal preference whether you like aliased images or not but sampling theory is not a preference.