alundeb wrote:
Peripheral illumination correction is something entirely different, and applied in software when decoding the RAW file. The sensor light falloff correction is applied before writing the RAW file, and is irreversible.
"The Peripheral Illumination Correction function has no effect on image data when the cameras are set for RAW, sRAW1 or sRAW2 recording modes, but the P.I.C. algorithm can be applied to these images during post-processing in compatible versions of DPP, together with additional image improvements including chromatic aberration correction and distortion control."
alundeb wrote:
Peripheral illumination correction is something entirely different, and applied in software when decoding the RAW file. The sensor light falloff correction is applied before writing the RAW file, and is irreversible.
"The Peripheral Illumination Correction function has no effect on image data when the cameras are set for RAW, sRAW1 or sRAW2 recording modes, but the P.I.C. algorithm can be applied to these images during post-processing in compatible versions of DPP, together with additional image improvements including chromatic aberration correction and distortion control."
Ruy
Ruy, I think you've misunderstood what's going on here.
There are two illumination adjustments that happen when an image is taken at very fast apertures. One is peripheral illumination correction, which is the compensation for vignetting. This is reversible because vignetting doesn't affect the center of the image, and Canon understands that automatic vignetting correction is not always desired.
The second adjustment, apparently, only occurs at apertures faster than f/2.8 and unlike vignetting, occurs uniformly across the entire frame. This adjustment is designed to compensate for light falloff caused by the physical depth of the sensor. At faster apertures, incident light reaches the sensor from a larger exit pupil and therefore from a larger incident angle. Microlenses can partially compensate for this issue, but unlike film, digital sensors without additional compensation fail to exhibit reciprocity at fast apertures. This second adjustment is designed to address this problem, and it occurs as part of the creation of the RAW file.
Canon documents the vignetting correction, but the second, which I don't have a name for (fast-aperture correction?) does not appear to be documented.
wickerprints wrote:
The thing is, if you didn't have this kind of compensation, then reciprocity would no longer hold. That is presumably why Canon implemented this scaling. Because the light falloff due to oblique incident rays is dependent on the particular sensor design, each sensor must have its falloff characteristics measured and a custom curve applied. Otherwise, two images of the same subject shot at the same exposure settings with two different cameras could be of different brightness.
Canon's sensors aren't the only ones that are limited by this physical property--it is just that they decide to compensate for it in the ADC in combination with the gain calculation for ISO. Yes, it does reduce DR but the better way to think about it is that you never had that DR to begin with--just like with ISO, shooting at a high sensitivity comes at a cost.
Now, I want to be absolutely clear that I don't necessarily like the approach that Canon uses here. I'm simply putting forth the facts that (1) explain their reasoning for doing it, and (2) show that such light falloff is not specific to Canon-made sensors....Show more →
As an engineer, this seems like a hack solution if there ever was one. I know they are trying not to offend the sensibilities of old film shooters, but this really should be resolved through shutter speed.
It would be interesting to see if you could mount both lenses and shoot them without having canon adjust them. The sigma would be easy because you could just turn it part way to make sure the contacts don't touch. And I think if you set the canon 85 aperture at 1.4 and hold down the DOF preview button while you remove the lens, it locks it at whatever aperture it was at when you removed it. Then if you only partially mounted it and took shots the camera would have no way of knowing what each lens was set at.
Right? Then you'd see if there was any actual illumination difference.
I agree that if they know this is happening then they should adjust the shutter speeds at larger apertures. But for people that shoot on Manual it would still result in underexposure.
SloPhoto wrote:
As an engineer, this seems like a hack solution if there ever was one. I know they are trying not to offend the sensibilities of old film shooters, but this really should be resolved through shutter speed.
I absolutely understand how you feel, but at the same time, adjusting the shutter speed to compensate is an approach I would be very reluctant to consider.
The reason why, again, is reciprocity. I am a strong believer in the importance of maintaining this mathematical relationship through as wide an EV range as possible, because it gives the photographer a consistent and predictable means of understanding the interaction between the components of exposure--shutter speed and f-number. It provides me with the ability to easily calculate how to change the exposure parameters to achieve the desired result.
On the other hand, I think it's also important that Canon would be more transparent about such things, and to at least give the savvy consumer the ability to make an informed choice and possibly more creative control. However, given the data I have seen I think this is a relatively minor effect.
Something else that is worth realizing is that whereas film is not afflicted with this issue (since the emulsion is receptive to light at virtually any angle), it is affected by its own form of reciprocity failure as it pertains to especially long (or short!) exposures. The intrinsic chemistry of silver halide is such that enough photons must strike the film to initiate a reaction. Furthermore, reciprocity failure in film occurs is wavelength-dependent. Finally, we speak of this fast-aperture issue for digital sensors as if this were a serious problem when the truth is, compared to the far superior ISO sensitivity of modern sensors over film, the effect is quite small.
So in the balance of things, I'll still gladly take the digital sensor. For all its flaws and limitations, CMOS/CCD has opened up far more creative possibilities for me, and film is not the perfect medium it is often nostalgically made out to be.
SloPhoto wrote:
As an engineer, this seems like a hack solution if there ever was one. I know they are trying not to offend the sensibilities of old film shooters, but this really should be resolved through shutter speed.
wickerprints wrote:
The reason why, again, is reciprocity. I am a strong believer in the importance of maintaining this mathematical relationship through as wide an EV range as possible, because it gives the photographer a consistent and predictable means of understanding the interaction between the components of exposure--shutter speed and f-number. It provides me with the ability to easily calculate how to change the exposure parameters to achieve the desired result.
I agree with wickerprints' reasoning. Ultimately, Canon does not design their SLRs for right-brained engineers and tech geeks. They design them for left-brained photographers and artists. It is important that the camera to maintain a consistent behavior.
I guarantee you a lot of those photographers would be dumbfounded if they dialed in 1 stop larger aperture, but received only 1/2 stop more light.
theSuede wrote:
Yes, this seems to be true for other brands also. But I think I can say for sure that only Canon keep this correction reserved for "only native brand lenses" - I can only test Nikon (D90, D700, D3x) and Sony (A850, A550, NEX5) right now, but it would seem that they correct for the given aperture regardless of lens brand - as long as it recognizes the lens communication scheme and can read the aperture setting.
BTW, the "NR" on Nikon raws is quite limited - it consists of a very simple pulse noise rejection scheme, a "speckle damper". It is not entirely defeatable. But neither is the quite more elaborate floating window median that Canon applies to all ISO1600 (and up) files that the cameras save... BEFORE sending the data through the in-camera settings of Noise reduction. :-)
This is getting totally out of scope for the thread - I just wanted to mention that the brightness difference between the Sigma 85/1.4 and the 85L isn't what it seems to be. The Sigma doesnt get the correction that the 85L gets. That's why the "difference" is visible.
The REAL difference is - just like I said earlier - about 1/5 Ev....Show more →
Wow this is enligtening. Thank you. That may also explain why Sigma lens works better on Nikon than Canon, who knows what else Canon is hiding...
We also need to take the lenses coating into this scenario. This is what happen between my 85L1 and 85L2. Open wide, MK1 is brighter than Mk2. I sold the Mk2 couple days ago so noway i can confirm the results but i am pretty sure the differences is about 1/4 stops last time i have checked.
mttran wrote:
We also need to take the lenses coating into this scenario. This is what happen between my 85L1 and 85L2. Open wide, MK1 is brighter than Mk2. I sold the Mk2 couple days ago so noway i can confirm the results but i am pretty sure the differences is about 1/4 stops last time i have checked.
If anything, newer AR coatings will be more transmissive and achieve higher contrast. Mk1 could be brighter for other reasons (like a marginal change in optics), but I'll bet against it being from its coatings.
In any case, whatever optical characteristics account for the difference between f-number and t-number, these would affect light-gathering ability uniformly across all aperture sizes. By contrast, the light loss at larger f-numbers due to sensor design is a function of f-number, with larger apertures resulting in a greater proportion of loss.
Again, with a concrete example: If a lens is stated to be f/1.4, then it may actually have a transmission of, say, t/1.6. We calculate this as a difference of
2 log(1.6/1.4)/log(2) = 0.385 stops
of light-gathering ability lost due to things like the fact that the glass is not 100% transparent. Then the same relative amount of light loss occurs regardless of whether the lens is wide open or stopped down. If the lens is shot at f/8, the equivalent t-stop is 0.385 stops slower; i.e., about t/9.1 (which is calculated as 8 x 1.6 / 1.4). In short, reciprocity still holds, even for t-stops.
However, in the case where light-gathering ability is affected by sensor well depth and the obliqueness of incident rays, clearly we see that for large apertures, each additional stop of opening up does not quite double the luminous intensity because the additional peripheral rays striking a given pixel become increasingly likely to be blocked from actually reaching the photodiode as the incident angle increases.
Wickerprints
Reading your last post brought to my mind the image of a cross-section of a lens, in which at least some of the elements are thicker toward their perimeter. Would this not result in less light passing through the thicker portion of the glass at larger apertures? I have no technical knowledge about optics; this concept just sprung up in my mind. It suggests to me that it may be that the t-number may differ from the lens's designated f-number more at larger apertures. Anything in that concept?
The amount of light lost due to absorbance is going to be pretty small. Most light loss is due to reflections off the glass surfaces, not from pure absorbance. Yes, the barrel will play a bigger portion of an effect at lower apertures, but probably not enough to notice. So t-stop vs f-stop should remain fairly constant across the board.
In the case we're speaking of, look at the pictures from this link to see how light coming in at very steep angles to the microlens (aka low f-stops) is not going to be well absorbed by the photodiode beneath it, but more likely to run into the electronics along the sides.
The amount of light lost due to absorbance is going to be pretty small. Most light loss is due to reflections off the glass surfaces, not from pure absorbance. Yes, the barrel will play a bigger portion of an effect at lower apertures, but probably not enough to notice. So t-stop vs f-stop should remain fairly constant across the board.
In the case we're speaking of, look at the pictures from this link to see how light coming in at very steep angles to the microlens (aka low f-stops) is not going to be well absorbed by the photodiode beneath it, but more likely to run into the electronics along the sides. ...Show more →
Light absorbance is a non issue in optical quality glass. As you say the light loss is due to partial reflection from the air-glass or glass-glass interfaces. Uncoated glass can lose 4% of the light due to reflection in air, and hence the reason optical coatings are so important. Modern coatings can reduce reflection loss to well under 1%, but even it it were say 0.5%, with a complex lens design and many interfaces at which light can be relfected, you can still end up losing a significant amount of light.
The ability to collect very oblique rays is a different matter again, but this will mainly manifest itself as vignetting and really is not about light transmission losses alluded to above, as that can occur even for a fully axial ray, and is a whole topic in itself.
I'd say we're in agreement. The only issue I might have with what you state is the ability to collect oblique rays manifesting as vignetting. The reason assumed herein by Joakim (theSuede) is that additional gain is needed to hide the effects of less effective light collection at higher numerical apertures (lower fstops), since the incident light cone is much much wider (and thus a greater percentage of the light is coming from oblique angles).
If your talking about oblique light hitting the barrel being mostly manifest as vignetting, well then we're once again in agreement.
And funny that you should work on ultrasounds. One of my advisors has several students working on contrast agents. I work on the optics side though.
wickerprints wrote:
I absolutely understand how you feel, but at the same time, adjusting the shutter speed to compensate is an approach I would be very reluctant to consider.
The reason why, again, is reciprocity. I am a strong believer in the importance of maintaining this mathematical relationship through as wide an EV range as possible, because it gives the photographer a consistent and predictable means of understanding the interaction between the components of exposure--shutter speed and f-number. It provides me with the ability to easily calculate how to change the exposure parameters to achieve the desired result.
On the other hand, I think it's also important that Canon would be more transparent about such things, and to at least give the savvy consumer the ability to make an informed choice and possibly more creative control. However, given the data I have seen I think this is a relatively minor effect.
Something else that is worth realizing is that whereas film is not afflicted with this issue (since the emulsion is receptive to light at virtually any angle), it is affected by its own form of reciprocity failure as it pertains to especially long (or short!) exposures. The intrinsic chemistry of silver halide is such that enough photons must strike the film to initiate a reaction. Furthermore, reciprocity failure in film occurs is wavelength-dependent. Finally, we speak of this fast-aperture issue for digital sensors as if this were a serious problem when the truth is, compared to the far superior ISO sensitivity of modern sensors over film, the effect is quite small.
So in the balance of things, I'll still gladly take the digital sensor. For all its flaws and limitations, CMOS/CCD has opened up far more creative possibilities for me, and film is not the perfect medium it is often nostalgically made out to be....Show more →
I agree with what you are saying, and even as I was composing that post, I was thinking about the different ways they could implement it.
Show weird shutter speeds = confused users
Hidden shutter speed changes = lots of side effects, worst for me would be around the sync speed
Higher gain = nobody notices
So I can easily see why canon chose what they chose, but if I could have another option, you can bet that I would choose it.
Ever since theSuede brought this issue to my attention, I've been turning it over in my mind and trying to think how it could be resolved. A 0.5-stop difference isn't much compared to, say, vignetting in the corners (e.g., the EF 24/1.4L II wide open is -4 EV in the corner!).
I think the only really effective way to deal with the issue is to redesign the sensor. Backside-illuminated sensors would go a long way to reducing the impact of light loss at fast apertures, but so far we have not seen these in any Canon or Nikon bodies. The technology seems to be most often used for tiny sensors in mobile phones and consumer P&S.
On the photographer's end, I think it should be possible under some circumstances to change your exposure settings to counteract the gain that is applied. For example, suppose your normal exposure would have been f/1.4 @ 1/100s @ ISO 200, and suppose the camera automatically applies a +0.33 EV gain due to the losses at the sensor, so in essence you are really shooting at ISO 250.
To counteract this, all you would do is shoot at ISO 160 and let the shutter speed decrease. Of course, I'm just using hypothetical numbers and the real situation is more complex because not all ISOs are "real"--some are actually just tweaks in the same multiplication factor as discussed in the linked article.
In practice, if you ETTR, you can probably get even better noise performance if you are able to sacrifice even more shutter speed and the scene DR permits it.
Ultimately, I haven't found this issue to be that critical and I absolutely love to shoot at fast apertures. From time to time I have wondered why I seem to get very slightly worse noise performance wide open, but until now I thought it was just a figment of my imagination.
theSuede wrote:
Yes, this seems to be true for other brands also. But I think I can say for sure that only Canon keep this correction reserved for "only native brand lenses" - I can only test Nikon (D90, D700, D3x) and Sony (A850, A550, NEX5) right now, but it would seem that they correct for the given aperture regardless of lens brand - as long as it recognizes the lens communication scheme and can read the aperture setting.
BTW, the "NR" on Nikon raws is quite limited - it consists of a very simple pulse noise rejection scheme, a "speckle damper". It is not entirely defeatable. But neither is the quite more elaborate floating window median that Canon applies to all ISO1600 (and up) files that the cameras save... BEFORE sending the data through the in-camera settings of Noise reduction. :-)
This is getting totally out of scope for the thread - I just wanted to mention that the brightness difference between the Sigma 85/1.4 and the 85L isn't what it seems to be. The Sigma doesnt get the correction that the 85L gets. That's why the "difference" is visible.
The REAL difference is - just like I said earlier - about 1/5 Ev....Show more →
