Mark III and ISO

[kevinbriggs]

I realize that this particular camera body has not even been released yet, but I

did have some initial questions as it pertains to ISO settings -- for anyone who

would like to hazard a guess:

 

I'm presently using the Canon 5D. From the information released thus far from

Canon, is the ability to shoot in high ISO settings with the Mark III radically

different from the recently released 5D?

 

As a wedding photographer, having the opportunity to shoot in very high ISO

settings with low noise would often be preferable to trying to utilize various

flash setups.

 

Thanks for any and all input!

[m_barbu1]

<p><i>is the ability to shoot in high ISO settings with the Mark III radically different from the recently released 5D?</i></p><p>Given that the 5D's "H" mode is equivalent to ISO 3200, and the 1D Mark III's is equivalent to ISO 6400, I would say yes. Did you read the <a href="http://www.dpreview.com/news/0702/07022208canoneos1dmarkiii.asp">press release</a>?</p>

[digitmstr]

Bette yet, did you read the white paper? It has all the info you need.

 

The ISO is not 'radically different'. It has an extra stop (5400) and, supposedly lower noise.

[kevinbriggs]

I suppose that the question that I am attempting to get at the heart of is one that is presently unanswerable as a result of the camera body not being released to the public:

 

Is the Digic III sensor (as opposed to Digic II sensor) going to make a remarkable/dramatic difference when it comes to noise values at these high ISO settings?

 

Kevin

 

And yes, I have skimmed through the white paper. Thanks again.

[bobatkins]

It appears so. From images posted so far and from what Canon have said, it does look like the 1D MkIII should have the best noise performance at high ISO settings of any EOS DSLR.

 

Whether it's due to Digic III (which is a processor, not a sensor) or to the sensor and electronics or the 14-bit A-D conversion doesn't really matter much. It's the final result that counts.

[danield]

I'm not an expert but I think that what's more interesting about the Mark III is the 14-bit capture instead of the 12-bit available in current cameras. This usually means both better noise performance and higher dynamic range... But of course this is all just speculation, let's wait for those reviews...

[digitmstr]

If I recall correctly Canon claims at least a one stop difference in terms of noise (i.e. 6400 ISO noise equivalent or less than the 3200 setting of previous Camon cameras). There are samples around the web and the results seem impressive indeed. Like you, and many others, I like to use available light as much as possible.

 

I am confident that since Canon (and others) are constantly working to reduce noise we'll have virtually noise-free images at ISO 6400 (and higher) in a not-so distant future.

[damian_tinsley]

I haven't got the literature saved, and I can't remember if the 6400 is a 'boosted' 3200 or a real 6400. Or for those who really know how the sensitivity works on these puppies, is that even a relevant issue?

 

The 'H' setting on my 20D (I think I recall) is just a 1600 with an automatic stop more of shutter speed applied and then adjusted in camera. Is this the same for the 6400 or is there an actual 2 stops extra ISO sensitivity available?

 

regards

 

Damian

[jim_larson1]

Damian: That is the REAL question. . isn't it?

 

Just like the 1/3 stops on the 30D and 5D are really software manipulation of data, rather than true 1/3 sensitivity adjustments of the sensor.

 

I am optimistic that the 1D-III has really made a noise control advancement. I eagerly await reviews.

[fourfa]

from the white paper, is sure sounds like 3200 is a "real" ISO and 6400 is a boosted "H" mode. Assuming it's up to 5D standards, one whole stop is a big improvement.

[steve_dunn2]

<p>Canon is not in the habit of saying whether the highest ISO on their bodies is actually implemented in the sensor hardware, or if the sensor tops out at the next ISO below that and the body simply shifts the bits to implement the highest ISO (and they also aren't in the habit of admitting that the 1/3-stop increments are also not supported by the sensor hardware). So I don't expect they will tell us this about the 1D III, either. Someone will surely test it and come up with an answer once the camera hits the street.</p>

[chrisjb]

But is the ISO 6400 real

 

What is real with digital ISO, am I wrong that the sensitivity of digital sensors remain constant eg:100ISO. and that an increase in ISO is simply an amplified signal, as similar to music you amplify the music and introduce things like white noise and then filter it out. the noise introduced in higher ISO is being improved with better in camera noise reduction software continously. so newer model won`t need to underexp and overdevelop in software.

[eric merrill]

It used to be that the fastest shutter speed was the most inaccurate one. If you wanted accuracy, you stayed away from the top one. The 2nd fastest was usually spot on.

 

Seems to me like the high and low ISOs are sort of like this. Is the 6400 ISO for real? Probably not. But that most likely means that the 3200 setting *is* for real this time. Only time in the field will tell for sure. Same way, once they start giving us 25 ISO, the 50 will most likely be good.

 

 

Eric

[alibek]

Aside from the noise level degradation - which is natural as we increase the ISO setting - boosted ISO does affect the DR.

As it goes out of specification, it doesn't matter whether it's an ISO 50 or ISO6400 - on some Nikons you lose in Dynamic Range even with a 'default' ISO 100 - since the spec. sensitivity starts with ISO200.

 

So Kevin, I think the 1D MkIII might really help you to do more for wedding photography. Hi-ISO with no boost plus fast-aperture lens ought to make a difference.

 

(Sometimes I complain on my 30D with the f4-5.6 lens... I wish it was a 1D Mk3 :)

 

Regards, Ali

[steve_dunn2]

<cite>What is real with digital ISO, am I wrong that the sensitivity of digital sensors remain constant eg:100ISO. and that an increase in ISO is simply an amplified signal, as similar to music you amplify the music and introduce things like white noise and then filter it out</cite>

 

<p>Pretty much, yup. The light-sensitive part of the sensor in a digital camera is not digital at all; it's analog, and its output is fed through an amplifier and then goes to an analog-to-digital converter. ISO is adjusted by controlling the gain of the amplifier, which of course amplifies any noise along with the rest of the signal.</p>

 

<p>But the issue with H on many (most? all?) EOS DSLRs is that H isn't implemented the same way. For instance, I have a 20D, which has 100-1600 plus H (3200). 100, 200, 400, 800, and 1600 all involve different gain in the amplifier. For H, however, the amplifier is set the same as for 1600. The metering system intentionally underexposes one stop, and then once the image has been digitized, it's multiplied by two to restore a correct exposure. You lose a bit of sensor depth as a result, and of course the noise is multiplied by two just as the actual image data are. L is implemented similarly on bodies which have it; it says 50, but the sensor can't actually do 50, so it's 100 and then once the data have been digitized they're divided by two, which explains why Canon warns that you lose dynamic range in L.</p>

 

<p>On at least some of the EOS DSLRs with 1/3-stop ISO increments, they're also done the same way. The amplifier's gain control only works in full-stop increments, so the body compensates at the front end by metering and at the back end by adjusting the image data after digitization. This leads to what might seem like a perverse result: a graph of signal-to-noise ratios takes a sawtooth form, with some lower ISOs having <em>more</em> noise than the next higher ISO. IIRC, each intermediate ISO is implemented using the closest full-stop ISO (e.g. 250 is 200 with a 1/3-stop digital pull; 320 is 400 with a 1/3-stop digital push), so 250 has slightly less noise than 200 while 320 is actually <em>noisier</em> than 400. (I know I've seen graphs demonstrating this, and I thought they were at dpreview, but I can't find them.)</p>

[gaetano catelli]

thanks for the explanation, Steve. in view of this, would there be better image quality to shoot one stop under at 1600 and adjust in Photoshop than to shoot in H?

[gaetano catelli]

btw, do i understand your post correctly that the basic difference between the increase in ISO in the 20D from, say, 800 to 1600 on the one hand and from 1600 to H (3200) on the other, is that the first amplifies the signal prior to digitization, and the second does so after digitization?

[fourfa]

<img src="http://andyradin.com/tests/5D%20noise%20vs%20ISO.jpg">

<p>

seemed relevant

[m_barbu1]

<p>Andy, it might be more relevant if there were more cluse as to what the chart meant. Are these luminance noise levels? Chroma? Both? What is the units for the vertical axis? What are the conditions upon which this data was gathered?</p>

 

<p>Personally, I'd be more inclined to trust the data from <a href="http://dpreview.com/">DP Review</a>, partly because it's more compoletely, partly because of the scientific processes they use to arrive at their data, and partly because I've validated their data in my real-world experience with three different EOS bodies.</p>

[fourfa]

right, sorry. The vertical axis is taken from the data in Photoshop's histogram window, set to RGB not luminance. The photos were long exposures of the lenscap - uniform black except for noise.

 

units are the sum of the 8-bit RGB luminosity. The absolute value is meaningless, since it varies with the RAW conversions setting used. But the same settings were applied to all images, therefore any change is entirely due to the ISO setting at the sensor.

 

This effect is also very small and near invisible in real world shooting. But when pushing shadows, as you must from time to time, I've been much happier with my 5D since avoiding in-between ISOs. your results may vary. since this experiment is so trivial to duplicate, I'd encourage you to try it yourself rather than take my word for it (or attack me for it).

[jim_larson1]

Andy;

 

Thanks for posting that graph.

[m_barbu1]

Andy, I understand the reasoning for the test, but I'm curious: by any chance, did you try this with anything other than pure black? If so, I'd be interested in those results, too.

[fourfa]

a fair point, it's on my list but I haven't gotten around to it yet. A real life scene with deep shadows, shot at all ISOs, cut and paste a crop of the shadow area to all the RAWs, analyze the crops in PS. Needs to be something with absolutely static lighting, so probably an indoor tungsten lit scene. If so it might be best done with a blue tungsten conversion filter, otherwise you'll probably just learn that, yes, the blue channel is real noisy under tungsten.

[m_barbu1]

Thanks for the tip - that I did not know. :-)