I think the same info is on the Canon USA Digital Learning Center, but that website is not the most easily navigated; I couldn't find it there quickly.
The colour "problem" is no doubt kept at a level that Canon feels is "well appropriate" for medium-high level cameras. The level they're at right now would not be a problem at all if just the transmittance curves were smooth (which they're definitely NOT), but that's a problem that can be solved.
That's a beautiful skintone example, Richard. There are actually some "artistical" uses for the compression/expansion of the yellow to red range of colours - sometimes it works to your advantage - sometimes it doesn't.
Let's take that argument a bit further:
Sometimes it's GOOD to "pull" some orange over to pure skintone - it reduces "blotchiness" in skin-colour. For models/subjects with good skin, you can get awesome smoothness straight from raw, with almost no PP.
-But as soon as you get a negative/raw that contains a "flat" surface gradient that extends into both endparts of the yellow/red-band you will run into problems - somewhere on this gradient you will get a steep "transition" edge, where hue turns from yellowish to reddish - very suddenly! In mixed lighting, typically indoor, this means that you WILL get a yellow/red "border" line somewhere in faces. Combined with the expanded red, that most of the time in most converters don't turn over to purple (transition red>blue is tilted very much towards blue, hues stay "red" well into areas where they should have turned bluish red) soon enough, you get the impression that "reds are too strong and too bright" after you have WB'd for the skintones.
The example Richard showed just above falls into the first category quite nicely - and it looks really good. But from an "accuracy" point of view, I'm quite sure that the lady in red's dress had more than one red hue in it, and that it was more blue-ish (if Richards profile is "normal" - otherwise maybe not!). Right now it's "hue-locked" - it has different levels of brightness, but a PERFECTLY uniform hue. And I doubt the skintones comparing all the people in the image really were that perfectly uniform from person to person in reality either... But of course this cannot detract from the fact that it LOOKS good! Which always is the end goal - you just have to do some informed self-training to get to know what you like/want and how to get there. And the easiest/most secure way to get what you want is if the originals you start from are accurate under as many circumstances as possible. That makes your PP-work uniform - you need less time spent on individual shots.
Very few cameras outside the MF-format except the Sony A900 can do this with any certainty nowadays though...
I see it as warm bright red, almost pure sRGB red with a tad more green than blue.
Regarding the DxO "Colour resolution", as measured in "bits".
This is a very hard metric to "wrap your head around". As DxO measures it, the difference is roughly two bits - but WHAT does that mean? In a real picture? This is very often the problem with DxO-figures - they're very correct most of the time (I've yet to find a mistake, or something that doesn't correlate with my own measurements) - but very unintuitive, not very easy to correlate with A PICTURE.
Those two bits (doesn't sound as anything much to care about, aye?) of noise-covariance based "hue resolution" means that the 5D can accurately discriminate FOUR times as many individual hues as the 5D2. This is quite a LOT considering that the 5D2 has got lower noise to "pollute" the interpolation engine with.
This is roughly proportional to my own measurements, that are based on another totally different metric - linear resolution when measured not in b/w, but with colour-difference. Adding in the 40% higher linear resolution of the 5D2, the difference "shrinks" to just two times as many individual hues with the 5D.
You might argue that the difference is of no importance, and you may be right. But the general TREND for Canon is to lower this every generation as resolution goes up. Unfortunately, colour resolution lost is lost forever, you don't get very much back with downsampling or averaging, as this is done after a further dilution of the colour accuracy - the interpolation engine in the raw-converter.
And, btw.... Doing as Richard does and actually grasping the importance of the use of a custom profile does more difference to average colour accuracy than the difference between any two new cameras on the market today.
A camera only has the ability to resolve hues; or not resolve hues. What CAN be resolved then has to be profiled into something we can see on the screen/print, and this constitutes 90% of what a "camera colour" is. RAW colour is both very wrong and very non-linear - it's not even anywhere near "accurate". The profile makes the impression.
thw2 wrote:
Sorry, I am not entirely convinced by Suede's postulate.
I just used DXOMark to compare the color sensitivity of Nikon D3X (24 MP FF) vs Canon 5D (12 MP FF) vs Canon 5D2 (21 MP FF). The performance ranking of the cameras goes as 5D > D3X > 5D2. But the difference is VERY SLIGHT.
Perhaps this has more to do with the RAW conversion program Suede uses?
On a slightly different but related note, Jeff Ascough has this to say about skin tones:
"Nikon approached me in June 2008 about using their gear. I took a D3 on loan shot two weddings with it, and sent it back; I didn't like the skin tones. At high iso they looked artificial and digital... As I understand it Canon use a different method of onboard NR to Nikon, which gives Nikon the edge in terms of actual noise but at the expense of skin tone at high iso."
This assumes we believe Jeff who also says: "Think about it for a minute, if I think a camera is crap but I say it is fantastic, then don't you think everyone will find out once they buy the camera? Then my reputation is down the toilet." ...Show more →
the 50D definitely falls apart on the DxO graphs though for color sensitivity though and it definitely seems to have a noticeably worse chroma characteristic
sometimes it seems like the deep shadows of the 5D2 have very mushy color
Thank you. I guess, the problem is how the better electronics of 7D will overcome sensor advantage of 5D mkI. I saw some tests in which 7D comes very, very close to the original 5D, as far as noise is concerned.
Electronics {process power and logic} do help.
But it dont override CMOS device physics.
The best pixel pitch is still around 7 microns and below 5.5 is problematic
with existing Bayer color scheems.
The 7D has CSFM of 77 and classic 5D is 99 and 5D II is 135 {see link below}.
The new 1DS MK IV is at 5.7 microns with what is called "Deep Well" Sensors. This is an attempt to gather more light via phase shifting technology as we used in phase shift photomasks for microlithography.
I need more info on this but so far dont see it as practical w the published design.
elfanucchi wrote:
Electronics {process power and logic} do help.
But it dont override CMOS device physics.
The best pixel pitch is still around 7 microns and below 5.5 is problematic
with existing Bayer color scheems.
The 7D has CSFM of 77 and classic 5D is 99 and 5D II is 135 {see link below}.
The new 1DS MK IV is at 5.7 microns with what is called "Deep Well" Sensors. This is an attempt to gather more light via phase shifting technology as we used in phase shift photomasks for microlithography.
I need more info on this but so far dont see it as practical w the published design.
Please quantify and justify your assertions. That and what on earth do you mean wrt phase shifting in the well? I know what it means in photolithography, but the principal isn't flowing over for me.
Please quantify and justify your assertions. That and what on earth do you mean wrt phase shifting in the well? I know what it means in photolithography, but the principal isn't flowing over for me.
My camera assertions are essentially the same as in Clark's report {link above}. It's lengthy ...
Phase Shift Photomask is what I did as the last 5 years of
Photomask Engineering Mgr. at Motorola.
Thats a very deep subject.
In short ,,, all image light is made of signal and noise and as it approaches the physical limits of optics and defraction, noise over powers signal or becomes too large of a componet.
There are various methods of phase shifting {ie. Alternating, Attenuated, etc} and all increase the signal by shifting the noise to the signal portion.
In Microlithography we have created sub wavelength structures with such soft and hard phaseshifting and that is why we can produce 60 Nanometer structuces using UV illumination that is 240 Nanometers in wavelength.
The reduction of gate and feature size is what the advancement of microelectronics in the 70's to 21st Century is all about. Phase shift since '99.
To see how it applies to the 1DS MkIV and called "Deep Well" by Canon see http://www.usa.canon.com/dlc/controller?act=GetArticleAct&articleID=3108.
Its in the IQ part of the presentation.
In short they hope to get more signal and less noise by cheating physics as we also did in microelectronics. They use a "phase shift" cover glass.
Just as a quick reply--I wasn't doubting you, I just still am a bit confused. Definitely not at the photomask development engineer level ;-) so I'm sure you know your stuff.
I got confused by your nomenclature--think I'm closer now though.
I've got a decent grasp of phase shifting in order to do subwavelength resolution lithography, but I'm struggling to see where in the stack they're using the phase shift glass in order to bend the light appropriately. Just beneath the IR cut filter? Is this in addition to the antialiasing filter, which serves to break the image up a little bit?
Nomeclature is the keystone of any engineering !
Kinda hard to define in simple terms unless you are in pool w wet feet.
See the 1DSMK IV link - its not specific on your stack reference.
Your reference to "Stack" is very much a microelectronis engineering term.
I would assume ... the phase shift cover glass replaces the normal passavation
layer used in any semiconductor manufacture.
Passavation is used as a last layer cover glass and protects the device from moisture and other factors.
It would be the Top of the semiconductor device {sensor} and
well under the AF / IR filters. Those filters are above and are seperate part of sensor.
I would be most interested in how they create the 'BUBBLES' on the passavation layer and you can be sure this is not intended to be public info.
This in effect would produce a weak phase shift effect.
I think "Deep Well" is an interesting appoach but ....
the real problem with commercial Foveon is light does not get into the stack - its a thick device.
I believe the US Govt microelectronics program has solved this problem w Foveon but thats my guess and its at least 10 years from the public.
Would also love to see real world images from "Deep Well" and "Normal Well".
Bayer MP count would be the "binned" or group number.
With Foveon ... "correctly engineered" ... you would have effective up to 4 x pixels.
That is how you get say 50 + MP in a FF DSLR.
How do you think the US military can read a newspaper from a hundred miles away in space ?? That kinda technology and a few more tricks no doubt ...
elfanucchi wrote:
Please quantify and justify your assertions. That and what on earth do you mean wrt phase shifting in the well? I know what it means in photolithography, but the principal isn't flowing over for me.
My camera assertions are essentially the same as in Clark's report {link above}. It's lengthy ...
Phase Shift Photomask is what I did as the last 5 years of
Photomask Engineering Mgr. at Motorola.
Thats a very deep subject.
In short ,,, all image light is made of signal and noise and as it approaches the physical limits of optics and defraction, noise over powers signal or becomes too large of a componet.
There are various methods of phase shifting {ie. Alternating, Attenuated, etc} and all increase the signal by shifting the noise to the signal portion.
In Microlithography we have created sub wavelength structures with such soft and hard phaseshifting and that is why we can produce 60 Nanometer structuces using UV illumination that is 240 Nanometers in wavelength.
The reduction of gate and feature size is what the advancement of microelectronics in the 70's to 21st Century is all about. Phase shift since '99.
To see how it applies to the 1DS MkIV and called "Deep Well" by Canon see http://www.usa.canon.com/dlc/controller?act=GetArticleAct&articleID=3108.
Its in the IQ part of the presentation.
In short they hope to get more signal and less noise by cheating physics as we also did in microelectronics. They use a "phase shift" cover glass. ...Show more →
Phase shift can refer to many things. In the context of cover glass, and specifically the AA filter, I believe there is a layer between the two birefringent layers to turn the linear polarized light coming through the first OLPF layer into circular polarization going into the second. This is done through shifting the relative phase of the two basis polarizations. Rather a different application from the quantum optics you mentioned in conjunction with semicoductor fabrication.
I didn't see any reference to "phase shift cover glass" in the video you linked to.
LoL - When non engineering professionals think they know and dont have the background to apply technology of "reading between the lines"... its time to unsubscribe.
Have fun 21st Century Digital Dogma Photographers ...
I really think Elfanoucchi is talking about the microlenses here--given that he's saying they replace the passivation layer (which is exactly where the microlenses sit).
Different terminology for sure, but the model fits if that's the way you're used to looking at it.
thedigitalbean wrote:
No, on an 18mp Bayer sensor there are 18 million photosites.
Is a photosite (photodiode) is the same thing as a pixel?
Using the 7D as an example, the math would suggest so (5184 x 3456) ~ 18mp.
From a color resolution perspective (even with interpolation), it's not quite the same as having RGB color information collected and represented by a single pixel (Foveon design). In fact, to me, Bayer sensor resolutions seem a little bit misleading.
? In a real picture? This is very often the problem with DxO-figures - they're very correct most of the time (I've yet to find a mistake, or something that doesn't correlate with my own measurements) - but very unintuitive, not very easy to correlate with A PICTURE.