Film vs Digital - Color Rendition

Discussion in 'Film and Processing' started by mauro_franic, Feb 26, 2011.

2. helgefrisenette

I swear there is line changes and paragraphs in the above! Something is awry with the forum software.

4. rodeo_joe|1

This thread was ridiculous nearly 8 years ago, and it's even more ridiculous now, since many of the images have disappeared from it.

- What side-by-side comparison image would that be? Give us a clue what on earth you're talking about, and why you've resurrected this thread.

- Well you're going to if you don't make any cogent point in the process.

This thread was severely lacking illustration in support of many of its assertions in 2011. Adding more words with no supporting images does nothing to strengthen or undermine any of the arguments previously put forward.

So what's your point; and what visual evidence can you supply in support of it?

5. glen_h

I don't know where this came from but ...

With three non-negative values, you can represent colors in a triangle on the CIE color diagram.

Less than perfect sensitizing dyes, dye-cloud dyes, Bayer filter dyes, or scanner filters will reduce the available color space.

In some cases, you can do a linear transformation (matrix) to compensate for such imperfections.
Mostly, you can do that if the spectrum is accurately known.

Scanners should do this if they have the information, or approximate it if not.

6. rodeo_joe|1

- Can I quickly put this old lame duck out if its misery?

Here's a photomicrograph of the dye clouds in a typical
100 ISO colour film at a mid-density.

To give an idea of scale; the smallest individual specks of colour are between 2 to 3 microns across. The clusters or chains of dye-clouds are on average about 10 microns or greater.

Now a 24 megapixel APS-C digital camera has a 'pixel' pitch of 4 microns, but we have to multiply that by 1.5 to put things on an equal footing with a 35mm film frame. That gives us an area of 6 microns square, which would enclose an easily countable number of dye clouds from the above sample. Maybe 10 to 12? That gives us a possible 10 or 12 levels each, of Cyan, Yellow and Magenta dye density.

So the number of possible colours in a 6x6 micron patch of that film would be (12x12x12) = 1728. And I think you'll agree that's being generous looking at the above.

The same, scaled down, 4x4 micron area of our digital sensor can represent (255x255x255) = 16581375 colours. Or over 9000 times as many shades of colour than the film, for a similar subject area.

That's a massive, massive difference. I could be out by two orders of magnitude in my estimate of film's colour-representing ability, and it would still lag well behind the poorest 8 bit/channel colour capacity of a digital sensor.

But wait! Any digital camera capable of RAW capture can hold 12 or 14 bits/channel of colour refinement. So make that a factor of 16 or 64 times greater still per channel in favour of the digital sensor.

That's a mind-boggling margin to lose by, and a pretty indisputable differential to explain away with any degree of credibility.

7. helgefrisenette

This is exactly what the internet and in this case online forums are good for. Sharing information across time and space.
WTF would or should I start a new thread when I can add to, find support in and add to posterity by posting in an old one?!
This is a thread that comes up often in searches of various kinds, and other people will find it in the future.

This is the picture I was referring to:

Try quoting my monoblock of a post above to get a more readable format. I don't know why the forumware does that?

8. helgefrisenette

Do you think you had luck in convincing anyone the last time you used that image (reverse image Google)?
You pull most of your numbers and assumptions out of thin air, while others are irrelevant.

One only has to look comparisons of real images to see that even if the scanner is not of the highest resolution, film wins most of the time WRT colour resolution and depth. There is the famous example of red berries on a bush, that is there on film but is lost between the gaps in the Bayer array.

You forget/neglect to mention that those 14 bits bit per channel are sampled from an analog sensor, with all the problems associated with that.
If there is much advantage to that kind of resolution, it will mostly be in avoiding beating and interference, both when sampling the analog signal and when doing various kinds og post processing, including basic demosaicing.
A full range of 14, 12 and I doubt even sometimes even 8 bit, is not present in the original signal.
Separation of colours with the dyes used in the sensors bayer array, is problematic too of course.

Film is not binary in any kind of way. There is naturally noise introduced in the signal by the grain of the emulsion, but there is no levels or quantification as such.

Last edited: Jan 6, 2019
9. rodeo_joe|1

- Of course it is. Learn something about photo-chemistry and then come back.

It's not me that's pulling figures and assumptions out of thin air, but you. Simply refuting facts is no argument, but I guess facts don't stand in the way of blind belief.

BTW, that picture of a map only reproduces the 4 CMYK ink colours used to produce the original map, and nowhere near enough shades of colour to prove or disprove anything.

Last edited: Jan 6, 2019
10. helgefrisenette

I know more about it than you assume. Tell me where and in what way is film photography, or the processing of it binary?
Simply dishing out numbers is not an argument either. There are few facts in your estimate above.

Regarding the map example. It is of course not an ideal test subject, but it is telling and ironic that is was used as a proof of digital superiority, while glaringly showing off the problems.
The missing colours are plain to see, but notice how the sharpness even in black elements is estimated/guessed. The stars are in fact distinguishable as stars on film albeit at lower contrast, while the digital version is just weird blobs.

11. rodeo_joe|1

- Please post a link to where that map image can be found. There's insufficient information to make any judgement as to colour accuracy.
To do that, we'd need the actual map in front of us.

- Only if we have some reference.

- In the case of a B&W emulsion, halide crystals are either reduced to opaque silver during development, or they're not. There's no shade of grey involved at all. The image consists of tiny specks of black silver on a near transparent ground. It truly is 'black and white' and quantised as such. On a microscopic scale there are no shades of grey to be found. I don't think you can get more binary than that!

The situation is similar for a colour emulsion, whereby tiny droplets of colour coupler are bound in an oily solvent and distributed throughout the gelatine carrier. Each droplet tends to be converted to dye, or not, in the presence or absence of developer oxidisation products.

Furthermore the 3 colour forming layers have a finite thickness, and can only hold a few dye cloud droplets within the depth of each layer. So in cross-section, you have those few dye clouds stacked to form a limited number of quantised levels of dye density. Not quite as binary as the B&W emulsion, but very limited in the number of levels of colour density that can be contained in a given area of film.

Can I ask if you have over 40 years experience of using film, and in some very technically demanding applications? Because I have. And such experience has given me a very good knowledge of film's strengths, weaknesses, and the processes involved.

Have you also examined film under a powerful microscope? I doubt it. Otherwise you wouldn't be making unfounded claims about its lack of quantisation.

12. glen_h

If all grains are the same size, then that is close to true. The result is a high contrast film.

But more usual, there is a range of grain sizes, where large ones are most likely to get enough photons to be developable.

The optical density increases as more grains develop in a given area.

With appropriate grain size range, a nice long straight (or close to straight) region on the characteristic curve,
results in the more usual contrast needed for usual lighting situations.

13. helgefrisenette

It's quite simple, some colours and features are doing a disappearing act from the film to the digital picture.
No known film artefacts can produce large features or anomalies like that.

The spongy structures AgX develops into are obviously there or not, but their extent is not equivalent to the original size of the crystal automatically. That is governed by the amount of exposure and development centers.
Also as mentioned, by Glen there is various amounts of overlap and size of the crystals to take into account.
As a starting point for extended reading and discussion I can recommend Tadaaki Tanis two excellent books.

Anyone can proclaim themselves as anything on forums. Also, you can easily be in a vocation for a lifetime without ever knowing more than what you need to know. You can easily get wrong mental models that persist because they "work" or are not an inconvenience.
I know university professors who think they are general experts in their field, but only really are reliable in their own specialty or vice versa.

Indeed I have, and also images of such are readily availible to googlers too.

14. rodeo_joe|1

- We don't know that for certain, since we don't have access to how those pictures of a map, or maps were taken. Without visual sight of the original, everything you're stating as fact is just assumption and guesswork. And it's just one example that you're refusing to give us a link to the source of.

There are thousands more examples online that clearly show the superiority of detail and colour rendering of digital, for a comparable sensor/film area. Such as here, and posted by a film user, but not a totally biased one.

Yours is the unscientific and biased approach, by refusing to look at evidence other than one single amateurish example that's apparently undocumented in its methodology.
- Another example of unsupported nonsense. Have a look at DXO labs analysis of digital sensor colour depth.
- It doesn't matter about grain size. The point is that reduced silver is opaque, and that dye clouds are fairly uniform in size and density.

What is seen in film is a dithering of opaque or coloured specks that only give the illusion of continuous tone.

It's exactly analogous to the half-tone printing method, whereby various sized dots of a single-colour ink give the same illusion. But the colour depth or refinement is limited by the smallest sized dot that can be produced.
- That has absolutely nothing to do with anything being discussed here. And as has been shown in another recent thread, what appears to be a straight line on a log/log scale is actually far from it.

Last edited: Jan 8, 2019
15. helgefrisenette

It's right here in this thread as I believe I mentioned earlier. Unless the guy who posted it for some odd reason fiddled with it to disproved his own point, I see no reason do disbelieve it.
Errors like that are typical of bayer demosaicing. They are often blatantly visible in certain details, but they profoundly affect image quality in the whole picture in subtle ways.

Did you even read the article/test you linked? It was very poorly and loosely executed, admitted by the authors themselves in the comments and they actually didn't conclude anything.

16. helgefrisenette

Edit: Sorry for the broken up post. Apparently I wrote something "inappropriate" or spam like that I wasn't able to purge, and I had to play divide and conquer.

Opaque but different in size and the dye follows the silver.

You could say something similar of just about any imaging technology.

Not exactly analogous. There is no raster and the "primitives" are can be approximated in shaped to the feature they are conveying. And of course the "dots" is very much smaller.

I don't see your point with the comparison?

Did you have a particular one in mind? Like this one: Nikon D850 : Measurements | DxOMark ?
There is still "only" 14 bits per colour channel. That is what the hardware can sample. As you can see the actual range is different.
No quantisation steps in film.

Last edited: Jan 8, 2019
17. rodeo_joe|1

- No, it doesn't. The dye couplers are activated by developer oxidation products, caused by reducing AgX crystals in close proximity to a coupler globule. The two aren't physically linked except by proximity.
- No, and you obviously don't want to.
Well how many bits colour depth do you expect from a 14 bit A/D converter? That's 2^42 different shades of colour and greyscale, and would require 4.398 x 10^12 separate pixels to display them all!

"No quantisation steps in film."

- Yeah, keep saying that and it's bound to be true.
I suppose that's why trying to change saturation or tone curves in a film scan rapidly results in visible posterisation?

Still waiting for the link to that dubious map comparison.

Last edited: Jan 8, 2019
18. helgefrisenette

That's it. Proximity and general volume. That is the factor that regulates the size of the dye clouds. And in turn their step-less size.

Look, we're not enemies, I'm not here to irritate you. If you have some real point to make I'd be curious to hear it.
The storage and displaying of the image captured by the analog sensor in a digital camera needs go through quantisation and locking into bit depth and a grid so in a sense if anything it's the digital image that is more akin to raster printing.
Silver halide imagings particular looks nearest cousin, would in my eyes be something like Floyd-Steinberg dithering

It's not the number bits themselves I'm pointing to (though they could easily be a source of artefacts too) It's the actual range and the distribution of the range.
IE, where and how are the quantisation steps placed? That's also a factor to consider.
The DxO test shows 22.6 bits at ISO 400. So that's under 8 bits per colour channel. Even at ISO 32 it's barely more than 8 bits of actual discerning ability per channel on average.

Because you are quantising the image when scanning it?

Real darkroom postersation relies the the upper and lower plateauing of film. In particular lith film and paper. That is still not a quantisation stepping though, since there is still a toe and shoulder, albeit a small one petering out into almost but not quite horizontal curves.

Really? You couldn't be bothered to scroll though the thread? Then let me do it for you:
Film vs Digital - Color Rendition first post by danielleetaylor