Film vs Digital - Dynamic Range

[bernardwest]

<i>If you & Vijay are claiming that the non-linear response on the upper end doesn't afford you much usable information, then how do you explain the presevation of highlight detail in negative film?</i><p>

 

I don't know (but take that with a grain of salt, as I am certainly no expert). I know virtually stuff all about sensitometry, and I am wondering what actually does happen after the shoulder. Vijay is suggesting that after the shoulder kink, the slope of the line zero's out. Is that the case? If it didn't zero out, but maintained a very small upward slope, then that could explain why adding exposure doesn't lead to a directly proportional increase in density. Certainly from the aspect of chemistry it seems to be a plausable explanation.

[mauro_franic]

Bernie, because I don't know. But both the actual results from my test and the Ektar publication (10^ -2.8 to 10^ ~3.1) show at least 14 stops. That is 50% more than the 40D. The 40D reaches the wall sooner, the exact explanation I'm not sure.

[rishij]

Actually I think we're all making this more complicated than it needs to be. Thanks Vijay for reminding us that the shape of

the curve doesn't have to have anything to do with maximal attainable dynamic range (though it can). Even within the linear

range of response, negative film can record a larger maximal vs minimal input signal. That says nothing about the max vs

min 'output' signal of the film, though, right? Correct me if I'm wrong Vijay. In other words, we could extend the dynamic

range of digital by having a photocell that can hold or record an even larger charge than whatever the 40D's photocells

currently can, yes?

 

I have a theory for the shoulder of negative film, along with some graphs from Baines... Will wait till I get home from the

airport to write more as this is just getting annoying on an iPhone...

[roger_smith4]

Okay, I am skipping some of the recent discussion as it doesn't appear related to the original one.

 

I downloaded the full size scan and raw file. I processed both with Lightroom and then used Photoshop to put them together as a jpeg (I had to stretch the digital image a bit to match the film's dimensions). No sharpening or noise reduction was applied to either file.

 

I attempted to roughly normalize the two scans using the white and gray patches from the QP Card and a bit of the black background so that there's some reference for comparison.

 

Doing that also meant that I couldn't completely optimize the raw scan to preserve highlight detail- increasing digital midtone values to match the film, retaining tonal relationships between highlights, midtones and shadows, and not blowing the highlights are mutually contradictory.

 

I'm not sure what this test shows other than negative film has a gradual roll-off of highlight detail but extreme highlights may not be color accurate (see what looks like a red cast). I don't see any scanner noise.<div></div>

[bernardwest]

<i>In other words, we could extend the dynamic range of digital by having a photocell that can hold or record an even larger charge than whatever the 40D's photocells currently can, yes?</i><p>

 

Just a bit of info related to this concept - the signal off a digital sensor usually doesn't occupy the full range possible (ie. in the 40D's case that would be 2^14-1 = 16383). I can't remember off hand whether it is the photocells that are filled to saturation, or if the ADC clips the results. I suspect it is the latter, because digital sensors (well, those from a year or two ago when I last researched this), are not fully linear. At the highlight end they can become non-linear. The clipping, I seem to remember, was to get rid of this behaviour (perhaps because the non-linearity was different depending on what colour channel you were talking about). In reality, the 40D only has values from 0 to about 13-odd thousand.

[vijay_nebhrajani]

OK, Arcane Post Alert. I'm going to try and answer some points raised, and may get into the really arcane stuff. Hopefully it is going to be interesting for some people, others please ignore as desired.

<p>

<i>Bernie: Vijay is suggesting that after the shoulder kink, the slope of the line zero's out. Is that the case? If it didn't zero out, but maintained a very small upward slope, then that could explain why adding exposure doesn't lead to a directly proportional increase in density. Certainly from the aspect of chemistry it seems to be a plausable explanation.</i>

<p>

Given that max density isn't infinite, there will come some point on the characteristic curve where more exposure won't do any good - all halide is already silver etc. Regardless of whether the slope after the shoulder is flat or slightly upward, a point must exist where the slope will become completely flat. That is your "max" point for dynamic range calculation. Same for the toe.

<p>

<i>Rishi: Actually I think we're all making this more complicated than it needs to be. </i>

<p>

Thank you. My point exactly. Dynamic range is a max/min ratio. No more, no less.

<p>

<i>Rishi: Correct me if I'm wrong Vijay. In other words, we could extend the dynamic range of digital by having a photocell that can hold or record an even larger charge than whatever the 40D's photocells currently can, yes? </i>

<p>

Oh yes indeed. In an electrical circuit dynamic range is limited by only two things (no surprise) - the noise floor and the peak signal output. For the active pixel sensor circuit I liked earlier, there is something called the Johnson noise (or Johnson-Nyquist noise) which is the noise floor. Below it, you can't distinguish signal from noise. The upper limit is a hard limit - no output signal can be larger than the supply voltage for the transistor amplifiers.

<p>

There is always a pressing need to reduce power consumption, so reducing the supply voltage is almost a given. The only ways left to increase dynamic range therefore are to lower the noise floor, or to do companding. Companding is easily implemented, but takes up area on the sensor die, area that should really be used by the photodiodes, reducing the "fill factor" for the chip. Thus the benefit that companding offers is negatively impacted by reduced pixel size, or greater noise. A possibility is to implement pseudo-companding; some people are trying that too.

<p>

There are ways to lower the noise floor - a simple one is to have larger pixels. Unfortunately, there is a limit to this too, since you want higher resolution (requiring smaller pixels) and greater dynamic range (requiring larger pixels).

<p>

Ultimately though, newer semiconductor technology will easily hit the 13 stop dynamic range of film (or 80 dB as it is called in the parlance) and possibly even exceed it to 120 dB (a 20 stop range!). I predict 13 stop DR for digital in 3 years or so, and 19-20 stops in about 5.

<p>

Hope I answered some questions, and hope it helps.

[juan_caballero]

Hey friends!

Don't worry about film or digital and just take good photos,some time you see people spend a lots of money in cameras and they take horrible photos,bad taste.I like this page but in general almost all photography are cheese and ugly,so you talk to much and spend to munch time in from the computer,go out and take good picture.

You can take good picture with any ship plastic camera, went you see a guy with big digital camera i almost sure that he take bad photos.

Thanks,Juan.

[marionovak]

wow, lots of confusion and lots of lack of knowledge,. here are few very simple answers,. Canon 40D uses CMOS

technology, and it is rather cheap camera,. while profesional systems such as Phase One ( I use P45+) have full 12

stop range,.. secondly,. scan should have been done on some top line scanner to compare those two,. juch as X5

from Hasseblad/Imacon and then you could see that in some ways digital back such as P45+ exceeds capabilities if

film, even when converted to b/w. I recetly posted few portraits. done digitally,. you are free to check<div></div>

[marionovak]

Above shot was done with existing lighting using PhaeOneP45+,.. no fill in lighting was used, pay attention to deatil,. both in the hightlights and in the shadow

[rishij]

<i>"wow, lots of confusion and lots of lack of knowledge"</i>

<p>

fair enough, but you contributed to reversing this 'lack of knowledge' how...?

<p>

p.s. i like this side of Vijay :)

[rishij]

To be more clear, Mario, if you think the digital back has more DR, any ideas fundamentally how? Higher bit A/D converter? Photocells capable of more charge?

 

These are the sorts of questions we've been trying to answer of late -- some of the more fundamental ones that might help us discern why we see certain behavior of film vs. digital when it comes to DR.

 

Sorry if Mauro's thread wasn't initially intended to go down this path... but it's entirely relevant and very interesting (for some).

 

Rishi

[rishij]

Vijay:<i>"Companding is easily implemented"</i>

<p>

How, in a digital sensor? Just curious if you have any ideas...

<p>

Rishi

[vijay_nebhrajani]

And just to add to my earlier points, take a look at these DP Review curves for the 1Ds Mk III - they all show

soft clipping; but the DR of this camera is barely 9 stops (8.6 stops).

<p>

<a href="http://www.dpreview.com/reviews/canoneos1dsmarkiii/page21.asp">EOS 1Ds Mk III curves</a>

[juan_caballero]

I forgot say that in some digital photos the image don't look real,is like Hollywood movies, fantastic, is to much,is because some people think like that in this days(plastic)

I like film just because look more realistic,like Hollywood movies from the 50' the good time in North-America.

Thanks.

[Dave Luttmann]

Nice shot Mario,

 

But silver based films can easily exceed 12 stops. No biggy there. I have obtained a very usable 14 to 15 stops with dilute solutions and stand development. Twelve stops is hardly a badge of honor for a $30,000 piece of equipment!

[marionovak]

ok,. sorry if I wasn't so clear,. if you compare film which is scanned with much higher quality scanner with CCD

sensor and then you compare that to Canon 40D which has CMOS sensor,. that test will be greatly in favour o film.

Also, development of technology is such that every few months new things come out.

 

What I was proposing was to test a film scanned on X5 Hasseblad scanner, which is about year old technology,.

versus some digital back like PhaseOne P45+ Both use CCD technology. I shoot both film and digital at the same

time,. film both color and b/w and I do about 7-8,000 shots on location each month, so I can freely say that I have

experience both scanning and processing digital.

 

Stil,. in some cases film is way to go,. but I found dynamic range of CCD digital back, pulling of the shots that film

will never be able to do. I am including one recent shot,. attention is to back lit windows on the sides, and the dark

face and white dress of Madonna at the middle of the shot,. only existing lighting i sbeing use<div></div>

[marionovak]

I agree Dave,. I have been shooting for 35 years and I am sad to see things like Rodinal dissaper,. which was developer of my choice,. I still shoot b/w,. changing back on my Mamiya allows me to do so,. and then have Velvia 50 for certain situation,.. which I scan with X5

[bernardwest]

Sorry, Mario, where is the 'confusion' and 'lack of knowledge'? You still haven't set us hicks straight on that one yet. Couple of points:

 

Firstly, you seem to be criticising Mauro for not using a P45+ and a hasseblad scanner. It would be great if we all just had one of these lying around, wouldn't it?

 

Secondly, whatever the p45 can do in relation to DR, negative film can obviously do it better. 12 stops is short of the 13-odd stops we are talking about for Ektar. So what's your point, other than a shameless plug for your own photography?

[jamesnjohnson]

I say someone use a Fujifilm S5 at 400% DR and compare it to Film.

 

Canon 40D is not the best of choices.

 

Wouldn't it be better to figure out which DSLR have the best DR first then compare it film, instead of using a DSLR that is not known for DR???

[mauro_franic]

James and Mario, can you please post a 14 stop graduated example like the mine so we can appreciate the dynamic range of your systems?

[marionovak]

Bernie, I wasn't taking about Mauro,. I meant lots of other posts that I have read following his,. so do I apologize to

Mauro,. I am not trying "to plug in" shamlesly my photography, I am way to busy proffesionaly for that. I was just

trying to explain to some other members not to mix apples and oranges and having engineering background, I know

how to set up accurate test. I was just trying to explain that when you do comparison,of two tests, film scanner and

camera should be in the similar price range, similar class,. sort of speak,. that gives more realistic test,. also that

both pieces of equipment are of similar generation. You can't scan film on CCD scaner and then compare it to CMOS

camera,.. that is my opinion,.. regarding P45+ and X5, it was just an example to explain myself,. two similar

technologies, that came out at about the same time

[marionovak]

Mauro,. from two images I have posted, and complexity of those lighting situation,. should be quite enough,.I believe photographers that dealt with similar situations will know what to look for. I do apologize if I have ofended you, I was trying to some other members explain difference in technologies and how aproach a test

[mauro_franic]

You have not offended. No worries. It is just not possible to see how many stops a scene had by looking at a picture of it.

 

Without any complicated setup. Could you take a one picture at f8 of a dummy metering f64 (with the meter to the camera) and post it?

[vijay_nebhrajani]

<i>Vijay:"Companding is easily implemented"

<p>

How, in a digital sensor? Just curious if you have any ideas...

</i>

<p>

Well, a sensor chip architecture is something like this: there are individual pixels; which are analog charge

accumulators whose charge varies with the photons striking it. This can be implemented with various

photoconductive materials (passive Light Dependent Resistors or LDRs) or photodiodes, or phototransistors etc,

which have a p-n semiconductor junction etc.

<p>

Then there is an amplifier that works as a buffer - a mechanism that allows the charge value to be detected -

i.e., read out, without destroying the original charge. It does deplete the charge slightly - but being an

amplifier, it takes a very small current (in picoamperes or less) and reproduces the same voltage as the

photodiode at its output. That picoampere (10^-12 ampere) current is allowed to flow for just a few nanoseconds;

this does not deplete the charge in the photosensitive device by any appreciable amount.

<p>

The way this is organized, an entire row is read out at the same time. This is the usual architecture of a

dynamic random access memory (a DRAM) - you provide a row select and all columns are simultaneously available to you.

<p>

At this point, the pixel values are fed into an analog to digital converter that converts the voltage read out

from a sensor to a digital representation. This is a process called quantization. There may be one ADC per

column, one for the entire chip, or any number in between. This is dependent on the analog to digital conversion

time, and how fast you want to read out the entire pixel array; and on the area of the die, power requirements

and so on. This affects the frames/second specification of the camera, so some appropriate number will be chosen

based on the overall data throughput that the sensor designer decides is appropriate.

<p>

The digital representation of the analog value can be 8 bits, 14 bits, 16 bits or whatever. This increases the

precision of the representation, not the dynamic range. The actual dynamic range is a voltage ratio, the "dynamic

range" of a 16-bit vs. a 32 bit representation is a meaningless number, since the max value - FFFF for 16-bit, or

FFFF_FFFF

for 32-bit represents the same voltage - the saturation voltage of the photo cell.

<p>

Once you have a digital representation, what you end up having is a matrix. As in a mathematical matrix of

numbers. You can apply several algorithms for post capture processing at both the hardware and firmware levels

and eventually get your JPEG or TIFF or whatever image.

<p>

With this background, you'll see that companding must be implemented at the level of the pixel itself - i.e., a

pixel must have a nonlinear response characteristic. There are semiconductor techniques that let you do this: see

this <a

href="http://ieeexplore.ieee.org/Xplore/login.jsp?url=/iel5/16/34442/01643509.pdf?isnumber=34442∏=JNL&arnumber=1643509&arSt=+1733&ared=+1735&arAuthor=Sungsik+Lee%3B+Kyounghoon+Yang">IEEE

paper</a>.

<p>

You could implement nonlinear amplifiers if needed, and <a

href="http://www.play-hookey.com/analog/logarithmic_amplifier.html">here</a> is a reference for this. These are

called "waveshaping" circuits, but I am not sure if these will help if the actual photosensitive element itself

has limited dynamic range.

<p>

I'm hearing of a new technique that converts light to frequency inpixel - this seems to give a linear dynamic

range of 115 dB and overall dynamic range of 130 dB (21.6 stops!!!). See this <a

href="http://ieeexplore.ieee.org/Xplore/login.jsp?url=/iel5/16/4016323/04016367.pdf?tp=&isnumber=&arnumber=4016367">IEEE

paper</a>. This is far more promising than putting companding circuits in the sensor - and it operates at a peak

voltage of 1.2 volts. Really impressive.

<p>

Finally, here is some excellent reading material if you want to understand more of the nitty gritty. <a

href="http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/">Noise, Dynamic Range and Bit Depth in Digital SLRs</a>.

[marionovak]

I will try,. I am on location on a big project,. how do you want to set it up? I don't have f64 on Mamiya,. it goes to f22,. but I can play aorund with the speed