Andrew J wrote:
"There is "technically correct" exposure, which occurs when a scene fits the sensor exactly and it is reproduced with detail from lightest to darkest content with texture (i.e., Zone 1 to Zone 9) and there is "perceptually correct" exposure, which can occur when exposure isn't technically correct but the viewer accepts it as real."
I love the way anyone can cut through the BS. The largest file size shot wins. The same scene with the same camera, the smaller file size shot is never going to be the winner. ETTR wins every time with a larger file size as proof.
Sorry, but there is no single "technically correct" exposure for a given subject or scene. There are a variety of factors to consider when you start looking at exposure this way, and one aspect of interpretation - a central and critical element of good photography - involves decisions about exposure.
There are useful principles that can be brought to bear on the question of how to expose a given scene with your intentions in mind. Simplistic notions like "largest file size wins" are based on some truths, but they are far from being effective operating principles.
winman3 wrote:
I posed a technical question thinking there is an absolute, empirical answer re ISO generated digital noise. Oh, gimme back the good ol' days of film !!!
Well, higher ISO setting does not increase noise, lower exposure does.
Andrew J wrote:
The largest file size shot wins. The same scene with the same camera, the smaller file size shot is never going to be the winner. ETTR wins every time with a larger file size as proof.
Not true. That's the same logic as saying all fat people are happy because Santa is always seen smiling.
RAW and JPG files vary in size according to the number of discrete colors and tones in the image. A shot of a blank wall will be smaller in size than one of a person standing in front of the wall because it contains more discrete tonal values and can't be compresses as much by the encoding algorithms. Noise artificially introduces more tonal / color variation, so if you shot an blank wall I would expect to see a slight increase file size wall as ISO increases.
Noise is entirely function of signal / noise ratio, specifically lack of signal in the shadows. How much of an image has visible noise in it is a function of sensor range, which in turn is a function of the physical volume of the sensor sites. Regardless of sensor design and DR all tones in an image have the same level of residual noise in them. But in lighter tones the signal masks it. By way of analogy you won't hear a 60Hz ground-loop hum in an audio system at high volume, only the low volume and silent pauses in the music. Also the more detail an image has the less likely noise will be visible. You are less likely to notice noise in a wide shot of fall foliage, compared to a black car shot at the same ISO in the same light.
DR and noise are related to sensor site physical volume: the capacity to accumulate photons and convert them into electrons at the individual R,G,B sites.
By way of analogy, imagine the sensor sites as buckets with the those in the highlights filled with a fire hose and the darkest shadows filled with an eyedropper. Exposure ends when the firehouse fills any bucket to the brim - the point of clipping the highlights. How much water the bucket in the shadows have a chance to collect depends on the volume of the buckets. The bigger the bucket, the longer it takes to fill them with the fire house in the highlights, giving the ones on the shadow more time to accumulate.
Color temp of the lighting affects noise. Tungsten light has very little blue energy in it relative to G and R. So even in the white highlights the blue sensor sensor sites will get less signal than the G and B sites. There are twice as many G and R and B sites because the sensor is designed to simulate the human eye which as 2% color sensing cone cells with the remaining 98% of the retina covered with rods which only respond to greenish light. To achieve WB in tungsten light the camera must amplify the signal from the blue sites more than R or G and as a result tungsten images have more noise that daylight or flash at the same illumination levels, which more noise in the blue channel than R and G.
As the sensor sites accumulate photons they are converted to electrons. There is a margin of variance in the efficiency of the sites in converting photons to electrons. When photographing a smooth surface like a wall all the sites should, in theory, output the same signal level. But in practice there is variation. The slight differences between RGB sensor output manifests itself and rainbow speckled noise. The more uniform in tone and color an object is in a photo, and the darker it is reproduced, the more you will notice the random variation and noise.
If there is no signal recorded in the shadows by the exposure, they should in theory be uniformly empty with zero charge and free of noise. But in practice when the sensor cells are 'dumped' to read the accumulated charge, 100% of the charge isn't released by the sensor site. There is always as residual voltage remaining like sludge like drops of water clinging to a bucket due to surface tension. This was illustrated in a white paper Canon a few years ago comparing its CMOS sensors to the CCD competition. So the while the shot you just took may not have recorded any signal in the shadows there may be a residual charge in the shadow sensor sites from the previous exposure which will create random variation in signal in the shadows which got little or no recordable light from the scene.
Sensor with larger pixels, generally speaking, have longer DRs. Longer DRs allow the sensor, when ETTR, to record more signal further into the darker shadows making the noise less visible because there are fewer shadow areas where no signal was recorded.
My 8MP 20D has larger sites and a slightly longer DR than my 15MP 50D, which means photographing the same scene with both should in theory produce a shorter range image with more shadow more noise in the 50D image. But as Canon packed more pixels into the 1.6 form factor it also tweeked the sensor site shape and volume and noise reduction in the A/D processing algorithms so the wind up looking very similar DR and noise-wise, with the 14-bit sensor of the 50D creating smoother tonal gradients and all things considered producing better IQ.
The latest development are "backlit" sensors. Backlit is a misnomer. There isn't a light panel behind the sensor uniformly flashing the cells as the name would seem to imply — which would help eliminate noise but shorten DR. The term refers to is the fact the circuits for reading the sensor which had always been placed on the top side of the cells has now been moved below them, on the "backside" of the chip. So compared to previous designs the sensor looks like it is being lit from the backside. Very confusing semantics, but that's true of most marketing hype. Moving the wiring to the backside allows using bigger buckets with thinner walls on the sensor, increasing volume for a given pixel density on the chip — the main factor for increasing DR and reducing the % of the an image where noise is likely to be seen.
Knowledge of how a sensor is designed and why it creates noise can be factored into ISO choices when shooting.
Detail: If shooting cars with smooth uniform surfaces noise will be more of a factor in ISO choice than if shooting a scenic shot with a lot of random detail.
Tone: A white car will be less likely to show noise than a black one as ISO is increase. They will have the same amount of noise, but on the white car it will be in smaller areas like the tires and wheel wells where is not noticed as much.
Color temp of the light: Tungsten lighting produces more noise in images than daylight or flash because it requires more amplification of the blue channel. So given the choice indoors I'll shoot with dual flash at ISO vs. ambient tungsten at ISO 800 so the images will have less noise.
gdanmitchell wrote:
You know what? In real photographic terms - say as observed in a 20" x 30" print - choosing the "native ISO" doesn't make a darn bit of difference.
Avoid the extended ISOs (like ISO 50). Generally use the lowest ISO that seems reasonable give the parameters within which you are shooting. You'll be fine.
I shoot a 5D2. If you gave me two versions of one of my better photographs, one shot at ISO 100 and the other shot at ISO 200, I couldn't tell you which was which.
Dan
+1
As an aside...I would like to see Canon refine IQ and WB than worrying about hi ISO. nearly 10years later in DSLR advancement and we still have unpredicatable WB with artificial light and too "red" reds...just sayin.
MrAdventure wrote:
As an aside...I would like to see Canon refine IQ and WB than worrying about hi ISO. nearly 10years later in DSLR advancement and we still have unpredicatable WB with artificial light and too "red" reds...just sayin.
Canon has and will no doubt continue to "refine IQ" as they introduce new cameras. To be honest, this is already far less of an issue with DSLRs than it ever was with film and it continues to get incrementally better.
"White balance" here doesn't quite make sense to me. WB is only relevant when you shoot jpg, where the WB settings have a significant effect on color balance and alter it in ways that are only controllable in the most general sense. When you shoot raw you more or less get the sensor data without these color balance alterations. If you mean "refine WB" in the sense of making better guesses about how to set it automatically in jpg mode, why not? But the problem cannot really be completely solved since the camera can only make generalizations about the overall scene.
If you mean "refine WB" in terms of raw... that doesn't quite compute.
If you are hoping for cameras to be able to analyze the scene with the same sophistication and judgment applied by skilled and sensitive post-processing experts... that will be a very tall order, indeed! I don't think that any manufacturer is going to be able to accomplish that any time soon.
As far as the "too red" (and the "unpredictable WB") issue goes, that is virtually always attributable to a small number of causes, assuming we are talking about raw mode:
1. Choices (or not making choices!) regarding how the raw file is converted that affect color balance in such ways, or failure to understand how to best convert and obtain the desired coloration.
2. Being accustomed to the way certain types of film handled color balance - and films varied wildly in this regard - and expecting a digital result by default to look like one's favorite film result.
3. The way that the image can "fail" with too hot of a signal in an individual color channel. The most common offender is the red channel. Knowing that response on digital sensors does not "roll off" the way it does on film, we can avoid this by shooting with the response of our sensors in mind.
gdanmitchell wrote:
"White balance" here doesn't quite make sense to me. WB is only relevant when you shoot jpg, where the WB settings have a significant effect on color balance and alter it in ways that are only controllable in the most general sense. When you shoot raw you more or less get the sensor data without these color balance alterations. If you mean "refine WB" in the sense of making better guesses about how to set it automatically in jpg mode, why not? But the problem cannot really be completely solved since the camera can only make generalizations about the overall scene. ...Show more →
Regardless of shooting RAW or JPG tungsten light has less blue energy than daylight or flash and the light reflecting off a white surface into the RGB sensor sites has much less signal in blue.
To achieve WB - render the white object as white - in tungsten light the blue channel must be amplified more than R and G. For JPG that amplification occurs in the camera. In RAW it occurs when the RAW file is displayed on the computer — ACR or whatever RAW editor you use amplifies blue/yellow channel in the RAW file more than it would in a shot taken in daylight.
Given the same shot in RAW and JPG the RAW will look better because the conversion process is more sophisticated in ACR vs. the camera
gdanmitchell wrote:
Canon has and will no doubt continue to "refine IQ" as they introduce new cameras. To be honest, this is already far less of an issue with DSLRs than it ever was with film and it continues to get incrementally better.
"White balance" here doesn't quite make sense to me. WB is only relevant when you shoot jpg, where the WB settings have a significant effect on color balance and alter it in ways that are only controllable in the most general sense. When you shoot raw you more or less get the sensor data without these color balance alterations. If you mean "refine WB" in the sense of making better guesses about how to set it automatically in jpg mode, why not? But the problem cannot really be completely solved since the camera can only make generalizations about the overall scene.
If you mean "refine WB" in terms of raw... that doesn't quite compute.
If you are hoping for cameras to be able to analyze the scene with the same sophistication and judgment applied by skilled and sensitive post-processing experts... that will be a very tall order, indeed! I don't think that any manufacturer is going to be able to accomplish that any time soon.
As far as the "too red" (and the "unpredictable WB") issue goes, that is virtually always attributable to a small number of causes, assuming we are talking about raw mode:
1. Choices (or not making choices!) regarding how the raw file is converted that affect color balance in such ways, or failure to understand how to best convert and obtain the desired coloration.
2. Being accustomed to the way certain types of film handled color balance - and films varied wildly in this regard - and expecting a digital result by default to look like one's favorite film result.
3. The way that the image can "fail" with too hot of a signal in an individual color channel. The most common offender is the red channel. Knowing that response on digital sensors does not "roll off" the way it does on film, we can avoid this by shooting with the response of our sensors in mind.
Thanks for that but you miss my point. Sometimes incremental improvement in a product line is better than cutting wide swaths with a machete. I'm sure there are plenty who like the wow factor of insane ISO numbers and machine gun FPS but I'm talking about items as I listed above. I get what your saying about "post-processing techniques/skill" to mediate those issues. I would like to see them addressed in the camera itself.
My personal camer is a 5D2 as well, no need to go into it's shortcomings here too much...but off the top of my head, the banding issues, WB, limited focus points could use some work...and maybe V3 will. The 1Dx looks to be a great machine on paper...just wished more effort was being placed in the smaller items.
cgardner wrote:
Regardless of shooting RAW or JPG tungsten light has less blue energy than daylight or flash and the light reflecting off a white surface into the RGB sensor sites has much less signal in blue.
To achieve WB - render the white object as white - in tungsten light the blue channel must be amplified more than R and G. For JPG that amplification occurs in the camera. In RAW it occurs when the RAW file is displayed on the computer — ACR or whatever RAW editor you use amplifies blue/yellow channel in the RAW file more than it would in a shot taken in daylight.
Given the same shot in RAW and JPG the RAW will look better because the conversion process is more sophisticated in ACR vs. the camera
cgardner wrote:
Regardless of shooting RAW or JPG tungsten light has less blue energy than daylight or flash and the light reflecting off a white surface into the RGB sensor sites has much less signal in blue.
To achieve WB - render the white object as white - in tungsten light the blue channel must be amplified more than R and G. For JPG that amplification occurs in the camera. In RAW it occurs when the RAW file is displayed on the computer — ACR or whatever RAW editor you use amplifies blue/yellow channel in the RAW file more than it would in a shot taken in daylight.
Given the same shot in RAW and JPG the RAW will look better because the conversion process is more sophisticated in ACR vs. the camera
That's right. As an interesting aside, the Nikon D3s's sensor was specifically designed to be optimized for this type of light (indoor color temperatures), to make its High ISO performance even better under such conditions.
snapsy wrote:
That's right. As an interesting aside, the Nikon D3s's sensor was specifically designed to be optimized for this type of light (indoor color temperatures), to make its High ISO performance even better under such conditions.
Ya...didn't want to introduce Nikon into the mix here but yes, the D3s that I use occasionally is somewhat better with hi ISO and artifical lighting OOC (JPEG).
Probably not a practical, nor cost worthy idea...
Perhaps a sensor optimized for low light, and a separate sensor optimized for high light?
And then for an extra added bonus, blend the results together?
The Ricoh model that is modular could do that now, though, would be expen$ive.
Just babbling out loud...
Though, I still would really like a model that has a TRUE native ISO 25, or ISO 50.
Not as a replacement for my current humble 5DMkII, but to compliment.
buggz2k wrote:
I still wish for TRUE low ISOs, 25 and 50.
wow, what an active thread this is. I will only contribute this to the original question. I met with a Canon "Explorer of Light" and his position was that for himself based on his experience, discussions with Canon engineers and others that the 'native' ISO for most Canon cameras was 160. That's where he shoots or multiples thereof rather than 100.
I can't and won't defend that, just passing on what he said.
Most Canon bodies can do ETTR but not many can do ETTL. I hate to constraint myself in such DR banding limitation. Imo, all Canon cam should support both way. Here is one sample of ETTL w/50L & 1Ds2 in wide DR of one summer.
buggz2k wrote:
Probably not a practical, nor cost worthy idea...
Perhaps a sensor optimized for low light, and a separate sensor optimized for high light?
And then for an extra added bonus, blend the results together?
Fuji did something along those lines with their superCCD -- it had smaller and less sensitive sensor sites blended in with the normal ones, to help extend the dynamic range.
When it comes to white balancing, making white white is not always the best approach - though there are some situations in which in can be the target. Let me offer one example and may be point to some related ones.
If you shoot snow under a blue sky, especially if the snow is in shadow, the camera "sees" the snow as being quite blue. In fact, it can be essentially the same blue as the sky itself. However, when your eyes/brain literally see this scene in person, a more complex perceptual process takes place. Roughly speaking, your brain "knows" that snow is white, so despite the objective evidence that this snow is the color of a blue sky you still see it as white.
So you might think that simply color balancing to make the snow white would be the answer. However, if you try that you'll likely discover that the correction went to far and that truly white snow - despite what your brain told you above - looks phony. So what is the right "white balance" for the snow? This turns out to be a largely subjective issue. It is likely somwhere between the sky blue that the camera recorded and the pure white that we imagine snow to be. Here a "correct" white balance will simply look wrong, and your "white" snow's color should be somewhere between actual white and the blue color that your camera registered.
To expand on this a bit. Let's say you photograph under a canopy of autumn leave because you like the effect of the warm coloration of the light. White balance that photograph and the very thing that attracted you to the light is removed from the image! Fail to make adjustments and the coloration in the photograph will seem more extreme that you perceived it to be while on location.
In your example of shooting under tungsten (or sodium vapor, or fluorescent, etc) light, you may well not want an "accurately" color balanced image. One thing we find attractive about subjects photographed under tungsten light is the warm quality it provides - i.e. we like the "inaccurate" color balance. If you color balance to the point that, once again, "white is white," then that warmth is lost. Yes, in certain types of photography that could be desirable, but in many others "correctly color balanced" would be undesirable.
Again, it isn't about finding the one "correct" way to do this. It is about understanding principles about color balance along with principles about perception and also understanding how applying these principles can obtain the particular aesthetic outcome you are after.
The idea that perfect white balance is something to be desired in all photographs is unfortunate. Now I can think of some situations in which such accuracy is perhaps necessary - for example certain types of commercial, advertising, fashion, etc photography in which consistent and objectively "correct" color is more critical.
Dan
cgardner wrote:
Regardless of shooting RAW or JPG tungsten light has less blue energy than daylight or flash and the light reflecting off a white surface into the RGB sensor sites has much less signal in blue.
To achieve WB - render the white object as white - in tungsten light the blue channel must be amplified more than R and G. For JPG that amplification occurs in the camera. In RAW it occurs when the RAW file is displayed on the computer — ACR or whatever RAW editor you use amplifies blue/yellow channel in the RAW file more than it would in a shot taken in daylight.
Given the same shot in RAW and JPG the RAW will look better because the conversion process is more sophisticated in ACR vs. the camera
tandlh wrote:
wow, what an active thread this is. I will only contribute this to the original question. I met with a Canon "Explorer of Light" and his position was that for himself based on his experience, discussions with Canon engineers and others that the 'native' ISO for most Canon cameras was 160. That's where he shoots or multiples thereof rather than 100.
I can't and won't defend that, just passing on what he said.
Ted
I've heard that same value bandied about. All I'll add is the observation that even that photographer would be unable to distinguish among prints made from originals shot at a few ISO values either side of 160 on the basis of noise or whatever other technical aspect you think might differ by some astonishingly small increment.
mttran wrote:
Most Canon bodies can do ETTR but not many can do ETTL. I hate to constraint myself in such DR banding limitation. Imo, all Canon cam should support both way. Here is one sample of ETTL w/50L & 1Ds2 in wide DR of one summer.
gdanmitchell wrote:
I've heard that same value bandied about. All I'll add is the observation that even that photographer would be unable to distinguish among prints made from originals shot at a few ISO values either side of 160 on the basis of noise or whatever other technical aspect you think might differ by some astonishingly small increment.
Dan
I agree with that...at normal viewing distances I'm hard pressed to see any difference between 100 and 160...matter of fact 200. I'm sure the mileage may vary under specific lighting conditions though.
Even for studio work the ISO 100 is the magic setting...I ramp that up to 160-400 alot with a huge distraction at normal viewing distances.
Which ever one comes out cleanest should be the native ISO ... Right? ... Or am I wrong?
Someone already did this with video on a 7D...cap on etc. I think 160 was the best in that case. I had the link once but no idea where I found it now. Search Vimeo.
