Nikon D7000 Dynamic Range

[ErikKaffehr]

Hi!

I don't understand the reasoning about ETTR. ETTR just says that we want maximum exposure without clipping non specular highlights. This essentially means that we are maximizing the number of photons detected. I can see there can be problems with the histogram representation, but would we have a "raw" histogram I would expect a correctly ETTR image to have maximum possible exposure.

Best regards
Erik

I agree that if one has a camera with low read noise and a high DR (such as the Nikon D7000), obtaining a perfect ETTR is not necessary since one might blow the highlights and lose data. However, the concept of ETTR implies that a histogram to the right is the critical factor, whereas in fact the exposure (number of photons collected) is the critical factor. Older cameras such as the Nikon D3 have higher read noise at low ISO, and when f/stop and shutter speed considerations limit exposure, it is advantageous to increase the ISO to take advantage of the lower read noise.

However, read noise approaches a minimum asymptotically at higher ISO, and raising the ISO beyond this point only serves to limit head room without decreasing read noise. As Emil Martinec has pointed out, this point of diminishing returns can be determined by looking at the DR plotted vs the ISO on DXO. In a camera whose DR is limited by shot noise, the plot becomes linear and DR drops by one stop for each doubling of ISO.  When read noise is significant, the curve bows downward towards the left, as shown for the D3 plot below ISO 800.

As the plot shown below indicates, the curve for the D3 becomes linear slightly below ISO 800. Increasing ISO above 800 with this camera limits highlight headroom and does nothing for the shadows; one can leave the camera setting at 800 and merely increase exposure in the raw converter. If the histogram is to the left at ISO 1600 and to the right at ISO 800, the DR will be similar, but with ISO 1600 there is a danger of blowing the highlights. In this sense, ETTR judged by the histogram is really a misnomer.

The D7000 is linear from the start, and one really does not have to increase the ISO to obtain a better read noise. One could leave the ISO setting at base and give as much exposure as f/stop and shutter speed concerns allow, and then increase exposure in the raw converter. Of course, the image will appear dark on the camera LCD, but highlight head room will be preserved.

Regards,

Bill

[Ray]

Hi!

I don't understand the reasoning about ETTR. ETTR just says that we want maximum exposure without clipping non specular highlights. This essentially means that we are maximizing the number of photons detected. I can see there can be problems with the histogram representation, but would we have a "raw" histogram I would expect a correctly ETTR image to have maximum possible exposure.

Best regards
Erik

Erik,
If ETTR essentially means maximising the number of photons detected (without clipping non-specular highlights) then ETTR is only possible at base ISO. Any exposure at a higher-than-base ISO, which does not result in a clipping of highlights, is an underexposure which has been amplified in-camera to a degree specified by the ISO setting.

To put it another way, any ISO setting above base ISO is an instruction to the camera to treat the signal as though it is an underexposure and to apply the appropriate degree of amplification in accordance with those instructions.

As you know, it's long been recognised with Canon and Nikon cameras that such amplification has advantages with regard to SNR because the amplification takes place before A/D conversion so that all further processing up-chain is effectively of a signal which is no longer underexposed.

Such an amplified signal will inevitably contain more noise than an ETTR exposure at base ISO, but the additional noise introduced as the signal is processed in-camera, starting with the A/D conversion, will have no more of an adverse effect on SNR than it would have had if the signal had been an ETTR at base ISO.

The consequence of this approach is that an ETTR exposure at ISO 200 has a better SNR than the same exposure at ISO 100, with most Canon and Nikon cameras.

The D7000 and Pentax K5 seem to have broken with this tradition. Unless you want a nice-looking review on the camera's LCD screen to show off the shot you've just taken of the nice-looking lady, it's probably better to underexpose 1 stop at base ISO than attempt an ETTR  at ISO 200, or underexpose 2 stops at ISO 100 than attempt an ETTR at ISO 400 etc, etc.

The advantage of this different technique with the D7000 is that there's no danger of blowing highlights but also no disadvantage of getting a worse SNR than one might have got using an ETTR at the appropriately higher ISO in relation to the same shutter speed and aperture.

[Ray]

But if the differences in noise are none between a D7000@1/50s and a 60D@1/25s and "barely perceptible" between a D7000@1/100s and a 60D@1/25s, would you not also expect the differences to be "barely perceptible" between a 60D@1/25s and a 60D@1/50s?

Not necessarily. If you double the number of photons collected, using the same camera, the noise in absolute terms can never be less. In fact it has to be at least marginally more. Photonic shot noise increases with increasing photon count, and read noise is also slightly greater with a greater signal.

What's important is the amount of noise in relation to the amount of signal, ie. signal-to-noise ratio. The D7000 produces less than half the noise in absolute terms as the D60 produces with twice the signal or twice the photon count, at base ISO.

[JR]

As you know, it's long been recognised with Canon and Nikon cameras that such amplification has advantages with regard to SNR because the amplification takes place before A/D conversion..
 

This is a sidetrack but....

Do you know if Canon`s DIGIC 4 processor is involved in removing noise from raw files? I know it is treating JPEG`S. Not sure about raw files.

In this article http://www.usa.canon.com/dlc/controller?act=GetArticleAct&articleID=1407 Canon is talking about their DIGIC 4 processor.

Quote from the section named "Canon Digital Photo Professional software" : However, while DPP can read the camera settings in effect at the time the images are taken, it tends to downplay the effect of the camera’s High ISO Noise Reduction. You may see little difference if you compare two RAW files, one with NR active, and one with it turned off.

Quote from summary: "It has a noticeable impact on the level of visible noise in shots taken at higher ISO levels, and can even have a positive effect on an image at lower ISOs....For RAW shooters, it can also make a noticeable difference in how Canon's software processes your images -- although with Digital Photo Professional software, you're free to independently adjust chrominance and luminance noise reduction beyond the initial in-camera settings."

I have never shot raw with NR turned on. From the article it seems like there is some work done to the raw file by the DIGIC 4 processor but perhaps it is so insignificant that it is difficult to see?


- John

[BernardLanguillier]

Do you know if Canon`s DIGIC 4 processor is involved in removing noise from raw files? I know it is treating JPEG`S. Not sure about raw files.

I believe that all cameras do some form of signal processing when converting the analog current coming out of the sensor to bits. That goes for DSLRs and digital backs.

I am not sure what the old myths is based on that some brands do less of that which results in better raw files. The good folks at Pentax where for sure explaining to me that the quality of their processing they apply to their files explains the very low amount of moire that can be seen in the 645D dng.

Let's face it, we don't really want to know what comes out of a sensor and real raw data isn't really available to anyone but the engineers developing cameras.

Cheers,
Bernard

[dosdan]

I recall reading somewhere that the further right one goes on the histogram the greater the number of photons one will collect.

Only at base ISO.

 
Dan.

[Ray]

Only at base ISO.

 
Dan.

That doesn't sound right, does it!

The further right one goes on the histogram, whatever the ISO setting, the more photons will be collected. However, if one wishes to maximise the number of photons the sensor is capable of collecting, one must be at base ISO.

[PierreVandevenne]

I am not sure what the old myths is based on that some brands do less of that which results in better raw files.

I think it is mostly based on very low light imaging, such as astonomical imaging. Canon's RAW files were much closer to what people expected from scientific based CCDs and yielded better results after a complete standard processing pipe-line. Nikon had a tendency of "eating" stars and dropping very faint signals (to cut a long story short). Of course, that type of imaging isn't totally congruent with normal photography needs where the goal is more "no visible noise in deep shadows" rather than "see these tiny details in the noise".

While reality is intrinsically noisy, we don't want to know about it and almost always prefer visual smoothness.

[Ray]

I think it is mostly based on very low light imaging, such as astonomical imaging. Canon's RAW files were much closer to what people expected from scientific based CCDs and yielded better results after a complete standard processing pipe-line. Nikon had a tendency of "eating" stars and dropping very faint signals (to cut a long story short). Of course, that type of imaging isn't totally congruent with normal photography needs where the goal is more "no visible noise in deep shadows" rather than "see these tiny details in the noise".

While reality is intrinsically noisy, we don't want to know about it and almost always prefer visual smoothness.

Yes, I believe that's the case. With the D7000 one can underexpose a shot, say at night, by 6 stops or more at base ISO, and when one reviews the image in ACR the screen can be totally black, initially.

One then hits the 'auto' button and the entire scene springs to life.

It's noisy of course, as one would expect any shot at ISO 6400 or 128,000 to be noisy, but the image still retains convincing blacks. There seems to be a clipping point for blacks which doesn't exist in Canon DSLR images.

The underexposed deep shadows in a Canon RAW  image seem to contain all the objectionable banding and pattern noise which gets progressively worse the deeper the shadow or the greater the underexposure.

Such an effect can be quite interesting, particularly from the 5D, if one is trying to create a semi-abstract photograph that looks like a tapestry woven from a coarse, reddish thread. 

[ErikKaffehr]

Hi,

I'd suggest that you are wrong on this. Increasing exposure will always increase the numbers of photons collected, independent of ISO.

But, we deserve better histograms (actually showing sensor signal and not processed data).

Best regards
Erik

Only at base ISO.

 
Dan.

[hjulenissen]

Hi,

I'd suggest that you are wrong on this. Increasing exposure will always increase the numbers of photons collected, independent of ISO.

But, we deserve better histograms (actually showing sensor signal and not processed data).

Best regards
Erik

If you keep everything else constant and increase the ISO, I think that the histogram will move to the right due to analog amplification, without changing the number of photons actually counted by the sensor?

-h

[LKaven]

If you keep everything else constant and increase the ISO, I think that the histogram will move to the right due to analog amplification, without changing the number of photons actually counted by the sensor?

I think this is what Erik is saying.  He'd like to see a histogram that reflects most accurately the numbers coming off the sensor.  As it is, we have to use things like UniWB to get the camera to approximate this.

[hjulenissen]

I think this is what Eric is saying.  He'd like to see a histogram that reflects most accurately the numbers coming off the sensor.  As it is, we have to use things like UniWB to get the camera to approximate this.

This is strange, indeed. If I use an audio-recorder, I expect its red/green LEDs to indicate clipping of its mic amplifier, A/D converter etc (usually, the clipping point of those should be close to each other, I think). I would not expect it to be tied in after automatic gain control, mp3 encoding etc. Why is photography any different?

My guess is that major camera manufacturers build their cameras mainly as jpeg cams. That is what a large part of their customers use, and it is easier to differentiate vs competitors on jpeg files than raw files. Therefore, the raw option is sort of an "added extra" for the nit-picking crowd.

I can see no technical reason why this is so difficult. What is UniWB, btw?

-h

[ErikKaffehr]

Hi!

http://www.guillermoluijk.com/tutorial/uniwb/index_en.htm

Best regards
Erik

I can see no technical reason why this is so difficult. What is UniWB, btw?

-h

[LKaven]

This is strange, indeed. If I use an audio-recorder, I expect its red/green LEDs to indicate clipping of its mic amplifier, A/D converter etc (usually, the clipping point of those should be close to each other, I think). I would not expect it to be tied in after automatic gain control, mp3 encoding etc. Why is photography any different?

My guess is that major camera manufacturers build their cameras mainly as jpeg cams. That is what a large part of their customers use, and it is easier to differentiate vs competitors on jpeg files than raw files.

I think you pretty much nailed it. 

In order to provide a preview on the camera's rear LCD, in most cases, the RAW data has to get passed through a JPG engine for rendering.  The histogram is generally derived from that. 

[ErikKaffehr]

Hi,

The RAW image itself is not really viewable, at minimum it needs to have a gamma curve, so some processing is needed, but I'd suggest that histogram should be based on raw data.

Anyone ride to see what happens if we overexpose at high ISO, do we get clipping?

Best regards
Erik

I think you pretty much nailed it. 

In order to provide a preview on the camera's rear LCD, in most cases, the RAW data has to get passed through a JPG engine for rendering.  The histogram is generally derived from that. 

[douglasf13]

Iliah Borg and Andrey Tverdokhleb contend that ETTR is mostly misguided with current DSLRs, and, once you start exposing midtones over a stop past midpoint, color starts to be negatively affected.

[stamper]

Iliah Borg and Andrey Tverdokhleb contend that ETTR is mostly misguided with current DSLRs, and, once you start exposing midtones over a stop past midpoint, color starts to be negatively affected.

Half of the information from the camera is in the stop at the right hand side. If that hasn't been captured then it is a bigger problem than the one outlined by yourself, assuming that is .... it is a problem? There isn't a definitive answer to capturing all of the information available so you have to judge what is the lesser of the evils and exposing to the right makes sense.