Push vs. ISO

[awinphoto]

Great advice as usual neuro. Noise will always be minimized by getting the brightest image you can get without blowing any highlights. One line of your response is what actually answers the OP...

But, if you've got the aperture and shutter speed you want for the shot, you're better off bumping up the ISO in the camera rather than in post.

But why is that the case? In other words, why does the camera do a better job of increasing the exposure (by ISO setting that is applied after the exposure) than adjusting the exposure in post (DPP, DxO, LR, etc.)?

I'm not going to pretend i'm an expert at sensors, neuro may provide a better explanation, but I would have to say it's because the camera/sensor is the first to see and process the file and can do what it needs to do with those beautiful digic processors and sensor to provide the best image possible... where as when you do it in post, the computer is pretty much blind in the essence that all it "see's" is code and information of the digital file that your camera CREATED... Then when you do heavy post to it, it's amplifying the information of the digital file but basically it's creating information where information wasn't there to begin with, so your left with the noise as a result... It didn't/doesn't know exactly what was there so it does it's best to guess for you. It was kinda the same in film... you could underexpose and push the film in development or overexpose and pull the film, but it never was quite the same as nailing your exposure the first time... and even at that, like digital, you didn't want to push/pull your film more than 1 stop max unless you wanted some funky effects.

Ok, yes the RAW file as written to the card may have been already manipulated/processed and that could be an explanation why an ISO increase is superior to a post exposure increase, essentially that's what I'm asking... I don't know much about what is written in a RAW file vs. what the actual photosite measured values are, is there some data that is lost/changed/hidden/discarded by the time the RAW file is written. But I'll disagree that the camera's processor (DIGIC for Canon) can do anything an Intel CPU can do. DIGIC is in part optimized for certain Digital Signal Processing directly in hardware whereas your computer CPU is truly general purpose and relies on software so it is possibly slower but not worse.

I'm not saying that the digic can do miracles the intel CPU cant, but what i am saying is the digic, at the time of capture, can have access to information that the CPU just cant, such as information about how much light is hitting the sensor, any extra tidbits of information the sensor can pump out in certain areas, etc... The computer in dark areas just recognizes dark areas with little to no information in the dark areas, hence when you try to then increase the lightness, the camera is essentially guessing at what should be there whereas the digic doesn't have to guess, it just reads the information from the sensor and processes it directly in real time... no guessing really needed.

 

[TexPhoto]

It's important to understand that when you bump up the ISO on the camera, it is not taking the same exposure and upping the exposure with it's processing. The camera increases the sensitivity of the sensor by adding more electricity before it shoots. Back in the film days you did this by swapping in ISO 400 instead of your usual ISO 100.

So bumping the ISO up and shooting is way different than shooting underexposed and pushing it up in post. Better? Yes, especially if you are talking about more than 1 stop. Why....

 

[Meh]

The camera increases the sensitivity of the sensor by adding more electricity before it shoots. Back in the film days you did this by swapping in ISO 400 instead of your usual ISO 100.

Sorry but not correct... at least not in the way you've stated it. The sensitivity of the sensors (the photodiodes) is independent of the ISO setting. The exposure is adjusted based on the ISO setting after the image sensor has been exposed. "Adding more electricity" is not a proper technical term but I suppose you could think of an amplifier that way.

Each photodiode in a CMOS sensor does have an amplifier to boost the signal as it's being read out but again this is after the sensor has been exposed and I'm not sure this is the amplification related to the ISO setting. This is not my field of engineering and I haven't come across any definitive information about how and where in the readout process the ISO gain is implemented. There may not be any one answer as there are a number of different implementations of CMOS and CCD image sensors.

 

[Meh]

Thanks but you're missing my point I think. Keying on your statement "you've got the aperture and shutter speed you want" I'm thinking of a situation where you just don't want or can't increase your exposure any further but the camera meter is telling you you're 2 stops underexposed. You've got two choices... take the shot and adjust the exposure in post or increase your ISO by two stops. This is, I think, what the OP is asking. You're advice was to go with the ISO increase. I'm asking why that's better than adjusting in post.

I did get the question, but sorry...I didn't answer it throroughly. Really, it depends on the ISO values in question. We think of ISO as gain, and that's true...but there are two kinds of gain, analog and digital. If all ISO changes were digital gain from a static base value off the sensor, then there would be no difference between changing ISO on the camera and changing it in post. For older digital cameras, I think that was the case. But, that's not the case now. Analog gain is applied in-camera up to a certain ISO value, which probably varies by camera/sensor/firmware, but is probably somewhere in the range of ISO 800 (Canon doesn't publish the data, but I recall some testing done on a 30D). Analog gain is applied directly to the RAW data before it's written to the RAW image file, and that's preferable to a digital gain.

"RAW data" implies a digitized value (i.e. post ADC) so "Analog gain applied to RAW data" seems contradictory unless you are using "RAW data" to mean the direct voltage values as read out from each photosite. There is definitely an amplifier for each photodiode but I don't know if this amplifier has adjustable gain depending on the ISO setting but your statement that there is some analog gain based on an ISO setting up to about 800 indicates that it is applied prior to the ADC. Anything after the ADC is obviously digital and includes all the readout noise in the digitally encoded luminance values for each pixel.

But back to the OP.... if you're right about the approximate ISO 800 being the upper limit for analog gain, does that mean the any ISO setting above 800 is digital and therefore no better in camera vs. in post processing?

 

[Meh]

I'm not saying that the digic can do miracles the intel CPU cant, but what i am saying is the digic, at the time of capture, can have access to information that the CPU just cant, such as information about how much light is hitting the sensor, any extra tidbits of information the sensor can pump out in certain areas, etc... The computer in dark areas just recognizes dark areas with little to no information in the dark areas, hence when you try to then increase the lightness, the camera is essentially guessing at what should be there whereas the digic doesn't have to guess, it just reads the information from the sensor and processes it directly in real time... no guessing really needed.

I did agree that the RAW file may have already been processed and information hidden/lost/etc. but by the time the camera's processor gets any data the output from the sensor (voltages) has already been through the ADC (A/D converter) and is nothing more than luminance values for each pixel. There are no extra tidbits of information... the photodiodes do nothing more than convert incident photons to electrons. The voltages are then read out and encoded into digital values (luminance levels) for each pixel.

I don't disagree (because I don't know) that the data is in a CR2 file has already been processed. I just don't know what data is present in a CR2 file, I assume that the luminance levels for each pixel (as well as other information about camera settings, etc.) are still present but perhaps I'm wrong about that?

 

[NotABunny]

Why does pushing in-camera ISO look better than exposure correction in post?

Most likely it's because the ISO is an analog process which is applied to raw sensor data (not to RAW digitized data).

If you take a photo, you a fixed number of tones (very few dark tones) to boost in post, which results in posterization (or large areas with the same tone - black or white).

But a higher ISO analogically amplifies the signal-noise and gets you the same number of tonal intervals at the corrected exposure (just like at the original exposure), albeit the noise makes the signal look more and more random (because the same signal gets distributed over a wider tonal range, that is, it's spread out more).

You can test this by simply trying to boost the exposure in post with, say, 4-5 stops (from ISO 100 to 1600-3200). Basically, you get garbage.

I don't believe that ISO 800 is any sort of analog limit in these times. For one, the new Cinema sensor has a native ISO of 850 (according to the Canon FAQ).

http://learn.usa.canon.com/resources/misc/cinemaEOS_faq.shtml

What is the camera's native ISO?
The EOS C300 has a native ISO of 850. It is at this setting that dynamic range, especially in highlight areas, reaches its maximum. At lower ISO settings, the dynamic range tends to shift more toward shadow detail.

Why 850 ISO?
ISO 850 on the EOS C300 provides an optimum balance between dynamic range and noise levels. At higher ISO or gain settings, dynamic range remains essentially the same, but digital noise tends to increase.

 

[Archangel72]

This is a little off subject, but I couldn't help myself not to notice one detail that Meh already point out.
One of our members (guess which one) have a beautiful 3 letters in his profile- 1Dx.
I hope we'll soon have a chance to read his 1Dx review.
Congrats

p.s. If you catch a little time, please share some information about your new "beast" (from practical aspect)...

Archangel72

 

[NotABunny]

p.s. If you catch a little time, please share some information about your new "beast" (from practical aspect)...

It's not his new camera, it's just the forum assigned "level". 1Dx will become public in march 2012.

 

[pedro]

What is the camera's native ISO?
The EOS C300 has a native ISO of 850. It is at this setting that dynamic range, especially in highlight areas, reaches its maximum. At lower ISO settings, the dynamic range tends to shift more toward shadow detail.

Why 850 ISO?
ISO 850 on the EOS C300 provides an optimum balance between dynamic range and noise levels. At higher ISO or gain settings, dynamic range remains essentially the same, but digital noise tends to increase.

@NotABunny: Thank you for posting this explanation. As I am not into tech at all this is an interesting revelation to me.

Asking a few additional questions then:

a)What is the current native ISO of a 5D2? (800?)

b)And what improvement might be expected in a 5D3? (1600?)

c) Is native ISO the more important base for qualifying a camera body noisewise: "this body in comparison to the other is one stop better in noise?" Or is it both components: sensor size PLUS native ISO?

Thanks for considering. Cheers, Pedro

 

[Archangel72]

p.s. If you catch a little time, please share some information about your new "beast" (from practical aspect)...

It's not his new camera, it's just the forum assigned "level". 1Dx will become public in march 2012.

I know that 1Dx will become available in March 2012. ... just... I hoped that (one of our members) became pre-tester of 1Dx, due to his great knowledge and activity here on canonrumors.
Sorry if I was wrong, no intention to cause any problems ...
My bad... :-\

 

[Edwin Herdman]

It should also be mentioned that some empirical testing (i.e. DxOMark's data, recompiled at sensorgen.info) demonstrates that benefits from increasing ISO start to decline, on any given sensor, past a certain point; this decline is accompanied by decreases in saturation capacity and dynamic range.

Therefore, I have said in the past (but honestly haven't found anything in real shooting to back up this claim, since I'm often trying to keep in line with "reasonable" exposures that look good on the camera's screen review, while avoiding ISO settings above 400 in most all cases) that there ought to be a point where increasing ISO on the camera makes less sense than pushing in post. But it might be the case that such a result will not be found as the camera manufacturers won't offer an ISO setting that is worse than pushing in post - at worst, this should only effect the most extreme ISO settings, and then it'd still need to be proven.

However, it ought to be considered that complex noise reduction processing is available for computer users that isn't possible in-camera, so quite often you should be able to get away with being a bit conservative or liberal with your ISO use, depending on the situation.

I'm no ETTR expert but if I recall correctly the main idea is to reduce the noise found in shadow areas, by exploiting unused headroom past the highlights. If that headroom was unused, shadow areas would be squeezed out.

 

[wockawocka]

It's well known that an incorrect digital exposure will create noise as a result.

If you are in a situation where you take a picture at the incorrect settings when you correct this in post production you will introduce more noise.

I'm not too sure how well this stands up if the ISO used was in the expanded range though as generally things get nasty up there anyway.

 

[GeorgeMaciver]

Wow, there is some photography expertise around here, for sure. I enjoy reading these posts, but don't often feel qualified enough to jump in. Thought I'd mention that I do love the forums though. Respect. Regarding ISO and cameras, I'm not sure if it's just me, but the 50D does seem to perform better around 400 to 500 ISO in most situations, though I've no idea why this would be, if indeed it is the case.

 

[awinphoto]

Meh, I did a quick search and this is what Canon says about the current digic 4 processor:

The performance of a digital camera is largely determined by three key components: the lens, the image sensor (CMOS or CCD), and the image processor. The image sensor, which takes the light entering the camera through the lens and converts it into electrical signals, can be compared to the film used in a traditional film camera, while the image processor "develops" the image. If any one of these three components is of substandard quality, it will negatively affect the images that the camera produces. Canon, which has been making cameras for over 70 years, develops its own lens units and CMOS sensors in-house, and has for some years now also been designing its own image processors: the DIGIC digital image processors.

The evolution of the DIGIC image processor has been the result of an ongoing search for higher and higher levels of image quality and faster speeds. DIGIC 4, which Canon began using in its camera products in the fall of 2008, represents the sixth generation of DIGIC processors. Compared with first-generation DIGIC processors, device performance has improved 50-fold in the space of just 10 years.
Today's DIGIC image processor does more than mere image processing: It controls a wide range of functions and circuits, including automatic exposure control, exposure mode control, image file compression/playback control, LCD display control, and more. All of these functions are contained in a single-chip system large-scale integrated circuit (LSI), and the same image-processing platform can be used in any Canon digital camera, including the Digital ELPH/IXUS and PowerShot series of compact digital cameras, and the EOS digital SLR (single-lens reflex) camera series.

One could say that DIGIC is the "brain" of Canon digital cameras. DIGIC 4, the latest version of the DIGIC digital image processor, provides dramatic improvements in terms of image quality and processing speed. So how have these advances in processor design been achieved? We talked to four members of the DIGIC 4 development team to find out more.

** On the digic5 press release I vaguely remember a key note at reduced noise... Canon claims in the above statement that the digic processes the image, hence controlling the noise... So it make perfect sense that the digic, if you take the statement above at face value, that the digic controls the noise better than the intel computer in this regards.

 

[neuroanatomist]

"RAW data" implies a digitized value (i.e. post ADC) so "Analog gain applied to RAW data" seems contradictory unless you are using "RAW data" to mean the direct voltage values as read out from each photosite. There is definitely an amplifier for each photodiode but I don't know if this amplifier has adjustable gain depending on the ISO setting but your statement that there is some analog gain based on an ISO setting up to about 800 indicates that it is applied prior to the ADC. Anything after the ADC is obviously digital and includes all the readout noise in the digitally encoded luminance values for each pixel.

TI guess I should have said 'raw signal coming off the sensor' vs. RAW data. As I understand it, there's an analog gain applied to the signal from the sensor before the ADC, up to a certain ISO setting. Above that threshold, additional digital gain is applied after the ADC.

 

[branden]

Does the analog gain vs digital gain correspond directly to the regular ISO numbers (100-6400) and then the "expanded" ISO (L, H1, H2)?

 

[neuroanatomist]

Does the analog gain vs digital gain correspond directly to the regular ISO numbers (100-6400) and then the "expanded" ISO (L, H1, H2)?

I don't think so. I think analog gain stops much lower than ISO 6400. But that's a guess, as I haven't seen data for the 5DII on this.

 

[Flake]

Just to add another factor and confuse everything a little bit more!

The other unmentioned issue is that of shutter speed, and all of this will be a little less accurate if shutter speed isn't considered. a 1/500 sec exposure and a Push, or high Iso is going to have very different results to a 5 Sec exposure. In most cases a 1 stop on the shutter speed will not have as large an effect as a push or Iso stop, but then that assumes that you're photographing a still subject with a tripod, if it's moving and you need fast speeds it's not of as much importance, but still needs consideration in the round.

 

[awinphoto]

Just to add another factor and confuse everything a little bit more!

The other unmentioned issue is that of shutter speed, and all of this will be a little less accurate if shutter speed isn't considered. a 1/500 sec exposure and a Push, or high Iso is going to have very different results to a 5 Sec exposure. In most cases a 1 stop on the shutter speed will not have as large an effect as a push or Iso stop, but then that assumes that you're photographing a still subject with a tripod, if it's moving and you need fast speeds it's not of as much importance, but still needs consideration in the round.

Are you referring to reciprocity and or long exposure noise referring to shutter speed? In the past with leaf shutters there used to be wide varaitions of speeds hence you had to have your leaf lenses with shutters built in tested regularly to insure your exposure was correct (wrong exposure could add noise especially if you had to correct in developing and or printing) but I'm just trying to get on the same page here....

 

[Meh]

** On the digic5 press release I vaguely remember a key note at reduced noise... Canon claims in the above statement that the digic processes the image, hence controlling the noise... So it make perfect sense that the digic, if you take the statement above at face value, that the digic controls the noise better than the intel computer in this regards.

Awinphoto, yes absolutely the DIGIC processor "processes" or "develops" the image from the original read out values coming out the sensor. It also creates the RAW file which is more than just the pixel values and is a proprietary format. However, in terms of processing or developing an image, I believe that has more to do with producing the JPEG rather than creating the RAW file. Good point about "if we believe" as certainly anything we read from Canon has marketing spin (not that they're lying but could be exaggerating).

The DIGIC processor only sees the output from the A/D converter... in other words it's already digitally encoded and therefore no more noise will be added (processing artifacts possibly, but again that's more about producing the JPEG). DIGIC5 is getting quite powerful but in general an in camera processor is subject to constraints (speed, power) that the computer is not and because of this some noise reduction algorithms, etc. that are too slow for in camera use could be used in post processing allow additional tools but not less.

Other than the fact CR2 is proprietary and Canon might know things about the data in the file that a 3rd party application would not be able to access, I doubt the DIGIC processor can do anything that can't be done on a computer and even then Canon would build that functionality in to DPP.