No, it's a construction optimization... :-)
The actual sensor surface only has a ~35-50% active light-registering ratio. The microlenses bring this back up to somewhere ~85-90% if you use F2.8 or higher lenses. But when you combine the two optical systems (lens + microlenses) you get an "overshoot" when the ML F-ratio is not matched to the lens F-ratio (or rather - rear pupil size, which isn't the same thing). You "overcorrect" the incoming rays, making them hit outside the active surface parts again.
For the majority of all users, this is a good thing. You get lower noise and better sharpness & less aliasing with everything at F2.8 or above. You lose some small amount at F2, and lose more at F1.4.
This is what generally "goes" with smaller pixels. But Canon isn't very good at this, the current lineup isn't that updated. Better sensors can get as low as 1/3 Ev loss at F1.4 - which is a very low price to pay for a 1Ev increase in light-efficiency usage average.
Ah, OK, that makes sense. Thanks Joakim for taking the time to explain these things - your posts are always very interesting and extremely enlightening
I have one final question, if you don't mind. We discussed earlier in this thread the color cast that wide angles produce when you put the exit pupil near the sensor. The standard answer is of course that rays with extreme angles partially miss the intended photo site and neighboring photo sites get activated as well (i.e cross-talk). The big unanswered question is why there is an asymmetry to the color cast:
i.e: Why is there a red cast on one side and a cyan cast on the other when the CFA and photo sites are symmetrical across the sensor and the lens is symmetrical as well.
denoir wrote:
In that case, why don't you apply the same principle when talking about lenses i.e a lens being sharp/contrasty relative the sensor size? An ASP-C sensor to produce an image with the same framing as an FF sensor would require that you use a wider lens. This subsequently will require a higher resolving power to get the same amount of detail. Very few lenses that I'm aware of are capable of producing the same contrast at 80 lp/mm as they can at 50 lp/mm.
The main point however is that I think it is an utter folly to start mixing lens and camera optics with post processed output. They are two separate things and should be treated as such. I think the 100 MP vs Rokkor example is a good one of how useless the DOF concept can be. As you pointed out the look and amount of blur was radically different yet according to the simple DOF theory they should have been close. In fact the Rokkor at f/1.2 should have had slightly less DOF, although in reality the 100 MP produced a whole lot more blur outside of the focal plane. ...Show more →
i guess i don't really care about dof as an optical concept, i only think about it when i'm taking a picture (what aperture/focal length for the look i want?). dof is meaningless without taking the camera and final picture into account. i do think about the transition from in focus to oof quite a bit though, as that contributes greatly to the individual look of the lens and is much less dependent on the camera (and that rokkor has a very unique transition). the flip side of that is that i almost never think about lens sharpness (though i definitely think about contrast and 'look' of a lens) when i'm taking a picture because it doesn't matter (most of what i do never gets printed large enough to see the differences in sharpness between my lenses).
denoir wrote:
(...)
What's the argument for using the same CoC? The same why we look at 100% crops when we want to see how sharp a lens is. Because resize and sharpen and other operations are bound to affect the final results significantly. We don't say that some lens is sharp on a certain format but not on another. Instead we have MTF charts that are a function of spatial resolution - i.e the MTF are not expressed as a function of line pairs but of line pairs per mm. (...)
Well, some of us actually mention lenses and their performance related to what format they are used with.
My old OM50/1.2 which made well on the 5D wasn't good enough when used with 4/3 cameras. "That's a good lens but will it be good enough for the small sensor?" I think I have heard that.
DPR, Photozone and SLRGear (like them or not) and... they all make different sets of resolution charts for different lenses.
Serious manufacturers give us MTF charts thinking of which format the lens is supposed to be used with. An Olympus ZD lens MTF chart is measured, or calculated perhaps) for 60 lp/mm where Canon MTF charts are for 30 lp/mm.
I think it is very hard finding an argument for using the same CoC for different formats.
denoir wrote:
The main point however is that I think it is an utter folly to start mixing lens and camera optics with post processed output. They are two separate things and should be treated as such. I think the 100 MP vs Rokkor example is a good one of how useless the DOF concept can be.
You're talking about blur, but DOF is about what is in sharp focus (at a certain display size/magnification) -- not about how blurred the OOF parts look. The definition is very narrow, but it is partly perceptual.
How much things are blurred seems to be hardly defined in optics at all, probably because of reasons along the lines of the thread that Tomser started about bokeh. I.e. blur probably isn't a very important factor in most fields of optical imaging (the function of a lens).
Jonas B wrote:
I think it is very hard finding an argument for using the same CoC for different formats.
I think that the designers of lenses would disagree with you. If we take Leica for instance, they provide DOF tables in the documentation that comes with the lens, and there's not a single mention of format or final image output or anything like that. They did not change their DOF tables when their flagship product changed from the M8 cropper to the M9 FF. Here's the f/0.95 Noctilux, look at pages 5-7.
AhamB wrote:
You're talking about blur, but DOF is about what is in sharp focus (at a certain display size/magnification) -- not about how blurred the OOF parts look. The definition is very narrow, but it is partly perceptual.
Yeah, that's why I said it was a rather useless term, at least if you want to describe the look a lens produces.
How much things are blurred seems to be hardly defined in optics at all, probably because of reasons along the lines of the thread that Tomser started about bokeh. I.e. blur probably isn't a very important factor in most fields of optical imaging (the function of a lens).
Zeiss has published a rather comprehensive white paper on the subject. I've just quickly glanced through it but it looks like it is worth reading. One interesting thing is that they talk about DOF quite a bit at first without any reference to format - just as a function of focal length, subject distance and aperture.
Afterwards they start discussing differences in format and circles of confusion.
Zeiss wrote
If we remove a lens from an old analogue camera and attach it to a digital camera of the same system that has a somewhat smaller APS-C sensor, then there is a "crop factor". We do not talk about an extension of the focal length, it doesn’t exist in this case. After all, the lens does not know how much of its image circle we are capturing with our sensor. The size of the object field is reduced by the crop factor while the object-side light cones remain the same, as long as we use the same lens and do not change the aperture setting. That is why the points of the light cones may not be located so far from the focal plane if we want to maintain the same ratio of diagonal to circle of confusion. Reducing the size of the film format therefore reduces the depth of field by the crop factor....Show more →
They repeat it a couple of times, that CoC is relevant if you want to keep track of scales.
Also:
Zeiss wrote
The depth of field is therefore a rather fuzzy dimension that depends heavily on the viewing conditions. Strictly speaking we can even find reasons for to use different circle of confusion sizes for different focal lengths of a camera.
sebboh wrote:
i guess i don't really care about dof as an optical concept, i only think about it when i'm taking a picture (what aperture/focal length for the look i want?). dof is meaningless without taking the camera and final picture into account. i do think about the transition from in focus to oof quite a bit though, as that contributes greatly to the individual look of the lens and is much less dependent on the camera (and that rokkor has a very unique transition). the flip side of that is that i almost never think about lens sharpness (though i definitely think about contrast and 'look' of a lens) when i'm taking a picture because it doesn't matter (most of what i do never gets printed large enough to see the differences in sharpness between my lenses)....Show more →
I guess I'll have to resort to saying: Yes damnit, it works in practice but not in theory!
No, but seriously, the more I read about it the more I'm convinced it's a too restricted and too fuzzy concept to really be useful. Even if you just consider the final image, do you always know what size you will be using or at what distance you'll be seeing it from? I mean, how many people actively think when they're about to take a photo.. hmm, I better change the aperture to f/4 from f/2 because I'll be posting this image on FM 1200px wide and therefor the DOF will be such and such to include so and so in focus.. I mean that is impossible and even worse, ridiculous. There are so many other factors at play that will far more affect the final image that it really doesn't make much sense in thinking in those terms.
denoir wrote:
I think that the designers of lenses would disagree with you. If we take Leica for instance, they provide DOF tables in the documentation that comes with the lens, and there's not a single mention of format or final image output or anything like that. They did not change their DOF tables when their flagship product changed from the M8 cropper to the M9 FF. Here's the f/0.95 Noctilux, look at pages 5-7.
CoC and lp/mm are connected, surely you can see that? We have sloppy manufacturers not caring about this when producing DOF tables but that doesn't mean we should lower ourselves to the same level.
A DOF table is from start something made up by conventions and assumptions including print size (enlarging) and viewing distances and vision, everything summed up as a certain needed CoC.
I also don't believe you really think a lens' DOF scale (a portable DOF table) stays correct when moving the lens from one format to another. For example, the DOF scale on the mentioned OM50/1.2 stayed the same whether I mounted the lens to my 5D or my Olympus DSLR. If anyone actually believed in that scale when using the lens with the Olympus DSLR they would be in for a surprise, and a disappointment.
I mention this as an example of how confusing things can be when not believing CoC is for real.
denoir wrote:
I have one final question, if you don't mind. We discussed earlier in this thread the color cast that wide angles produce when you put the exit pupil near the sensor. The standard answer is of course that rays with extreme angles partially miss the intended photo site and neighboring photo sites get activated as well (i.e cross-talk). The big unanswered question is why there is an asymmetry to the color cast:
i.e: Why is there a red cast on one side and a cyan cast on the other when the CFA and photo sites are symmetrical across the sensor and the lens is symmetrical as well.
Just an uninformed guess : sensor sample variation of mass-produced cameras might play a role, and insufficient correction by the firmware / in-camera processor.
One of the reasons you pay insanely high prices for MF digital backs is quality control, and advanced R&D and support for state of the art firmware.
denoir wrote:
I guess I'll have to resort to saying: Yes damnit, it works in practice but not in theory!
No, but seriously, the more I read about it the more I'm convinced it's a too restricted and too fuzzy concept to really be useful.
it's not, it just gives you a general idea of what the output PRINT will look like, but it is only accurate for a theoretical single element lens as far as i know.
denoir wrote:
Even if you just consider the final image, do you always know what size you will be using or at what distance you'll be seeing it from? I mean, how many people actively think when they're about to take a photo.. hmm, I better change the aperture to f/4 from f/2 because I'll be posting this image on FM 1200px wide and therefor the DOF will be such and such to include so and so in focus.. I mean that is impossible and even worse, ridiculous. There are so many other factors at play that will far more affect the final image that it really doesn't make much sense in thinking in those terms. ...Show more →
changing viewing distance doesn't make much difference until you get to the extremes same goes for print sizes. in theory things change but you don't notice it enlarging a size or stepping foreward or back a step or two.
since i shoot exclusively with liveview cameras, i only use it to estimate aperture setting up the shot, i look at the lcd (unzoomed for limited dof shots) to fine tune aperture to exactly what i want. this means i am theoretically optimizing for a print size of 3" diagonal (lcd size). it still looks the way i want at 12"x18" though. for landscapes where i want everything in focus i actually do zoom in to make sure everything i want sharp is sharp.
Tomser wrote:
Just an uninformed guess : sensor sample variation of mass-produced cameras might play a role, and insufficient correction by the firmware / in-camera processor.
One of the reasons you pay insanely high prices for MF digital backs is quality control, and advanced R&D and support for state of the art firmware.
No theories I'm afraid but I wonder if a review of the unconverted raw file of the C vs 3 files would help. I recall seeing some software that can show the raw values by colour and pixel but I can't for the life of me find it.
By box I meant what on an slr is called the mirror box. Unfortunately it doesn't look like it is any bigger on the c3.
Mike
masimo wrote:
Is the box into which the lens fits any bigger in the nex-c3 than it was in the older nex models. just checked nex-c3 images on google and there is not a single pic of the camera with the lens off. Not even dpreview saw fit to do one.
Maybe you could post a pic of your camera showing the inside of the box. It could be made large enough to avoid having to amputate the ears on the Contax G 21mm. In fact I wonder if modifying that box might be a better solution than trimming the protectors on the lens.
denoir wrote:
I think that the designers of lenses would disagree with you. If we take Leica for instance, they provide DOF tables in the documentation that comes with the lens, and there's not a single mention of format or final image output or anything like that. They did not change their DOF tables when their flagship product changed from the M8 cropper to the M9 FF. Here's the f/0.95 Noctilux, look at pages 5-7.
I think Leica skipped the whole crop mess for their lens designs, and always designed everything for full frame, specs, literature, everything. For Leica, FF was always the destination, and the M8 only a necessary interlude.
That cast is caused by NEX not having off-set lenses . Leica has same issue (different colors), but you wont see that cause they have nice in-camera build fix for that.
Its simply like in medium format, lens are too close to sensor for light to have proper angle to sensor in corners. Solution is probably bigger lenses or off-set lens on sensor itself. NEX 7 should have that..
Btw. does it appear only on lenses shorter than 28mm or even with lets say 50mm?
Tomser:
I'm afraid it's not sample variation as it is consistent.
Mescalamba:
The color cast shot is from an M9. Sony seems to have solved the problem with the NEX-C3. The cause of the color cast is not a mystery. What however has not been explained is why it is asymmetric (red on one side cyan on another) when every component involved CFA, photosites & lens are symmetric.
To answer your question, on the M9 it appears on 35mm and shorter. On the NEX-C3 it doesn't appear at all (or almost not at all).
denoir wrote:
...The color cast shot is from an M9. Sony seems to have solved the problem with the NEX-C3. The cause of the color cast is not a mystery. What however has not been explained is why it is asymmetric (red on one side cyan on another) when every component involved CFA, photosites & lens are symmetric.
I know of several posters, including myself, that has stated that the Bayer mask on the M9 is asymmetrical. Another poster, then, presented his proof that it was symmetrical, and apparently you, Luka, liked that explaination.
Hmm, thanks for that link. I wonder what they meant by it. I mean, what could possibly be the point of having the left side of a sensor being different from the right side?
The Bayer mask looks like this and is symmetrical:
I know the M9 Bayer grid looks like that because of an article in Leica Fotografie International (the official Leica magazine) that had a picture with explanations of how a sensor in a digital M worked. (There was no mention of non-uniform color casts).
I think also Ajay who works with sensor design pointed out that anything but a symmetric solution would be prohibitively expensive.
What's your reason for thinking that the sensor would be asymmetrical? What purpose would that serve?
I'm not too sure about the interpretation of that msg on the "Leica-FAQ" board. I'm fairly sure that the Bayer pattern is assymetrical, in the sense that the red "clots" of absorption dye are slightly larger than the blue&green ones. They're still "square" as in equal width and height, but slightly larger.
The sensor surface underneath however, isn't rotationally symmetrical. The ccd vias are slightly rectangular, the Y dimension of the active pixel surface being larger than the X dimension. This is the same in all of the Kodak 6.8µm sensors, dating back to the "original" in the Oly E-1, 2002. They try to mitigate some of this effect with ITO gates and metal masks on top of the sensor (to stop some of the colour crosstalk via pixel>pixel light spread), but there's also the question of angle/wavelength imbalance.
If you balance the QE so that you have one "correct" whitebalance for perfectly perpendicular incident angles, the Kodak construction will by construction default unfortunately not have the same WB for incident light with lower angles. If you shine a white light source with perfect collimation (all light travels in perfect, parallel lines in space) on the sensor - first at 90º angle, and then at 30º angle - you wouldn't get the same WB result... Even though the light has the same CCT in both cases.
This defect is aggregated by the use of quite thick (not the best materials...) colour filters and assymetrical cell structures (the blue cell is even more assymetrical than the other cells, red is almost square) and you get angle-dependent colour faults.
PREDICTABLE colour shifts, if you know the lens chacteristics, but even so.
I didn't say why the fault in itself would be asymmetrical over the surface, sorry.
To get the filter layer to work properly with light with high incident angles, the "normal" case you need to deposit them fairly centered over the active pixel areas. Unfortunately this means that they are DE-centered in the case of light with lower incident angles, due to the assymetrical nature of the CCD cell construction Kodak uses.
This decenterig means that you get one dominant colour fault in one direction from the image center, and the opposite colour fault in the opposite direction. And since the blue channel has the smallest active pixel surface, it has the lowest asymmetry. R-G shows a lot more deviation from corner>opposite corner.
[edit]
Note that the blue channel has the least vignetting.... (??!!) :-)
This is because the effective area used to collect light is slightly smaller in the blue cells. It has less leakage to/from the other cells too, for the same reason... The red and the green get more vignetting, and more leakage.
Hmm, but rectangular geometry would still be mirror symmetric, wouldn't it? In the images you have a diagonal asymmetry with the bottom left having a red cast and the top right a cyan cast. The cast is not perfectly symmetric either, but the variation there could be explained as variation in lens, mount and sensor alignment. The overall directions of the casts are always the same - you never get a cyan cast in the lower left corner for instance.
Edit: Sorry, didn't see your second post. I'll read it now.
