I think film grain is binary. Why do I think that? I learned it here:
Photo Utopia: Clumps and Chumps (or why film isn't binary)
The author meant to have the opposite effect, but after I read his explanation for why film is not binary, I formed a differnt impression.
I have been thinking a lot about film grain recently. I'm interested in mimicking its look in the digital domain.
I don't like the usual 'soft light or overlay some gray-centered noise' approach to making digital film grain, because I feel it doesn't really model what film grain actually is. That technique does not cause edge breakup, for example.
So What Exactly is Going On?
I needed to know if film grain got more or less gray (true analog), larger and smaller (like a traditional halftone, but with jittered cell placement), or denser and sparser (like a stochastic frequency-modulated halftone. In researching my questions I learned that this is apparently a point of some contention.
After mulling it over for a while, I decided I would go with this guy:
Clumps and Chumps or Why Bumble Bees Can't Fly and not just because he knows about bumble bee wing vortex shedding, although that did give me some added confidence :D
So here is how I think it goes, based on what I've read here on teh internets.
I think after development it's little black crystals of reasonably similar size that scatter and overlap. The more light a region gets, the more numerous the crystals become.
I think a negative image (or 3 color separations combined as here) would make a good start. Maybe stochastic halftones would form a good basis. Could the halftone be 'crystallized' using, maybe, the median filter? I wasn't thrilled with the results, but another method for crystallizing the halftone might be good.
I saw a quote on one of the preceding links from a Kodak scientist who flat out said film is analog, but Kodak needs to hold to the line that film is superior to digital, so I don't put that much weight on that statement. (Film is superior to digital in many ways, btw, and what's motivating me in this post is my desire to make digital seem to be more filmic). Plus at the right scale an FM screen will seem analog.
That same link concludes (ironically) with this quote:
So next time some 'expert' spouts off about bumblebees and how grain is binary tell them it's about "the energy of an electron being raised into the conduction band from the valence band"
I say 'ironic' because in quantum mechanics there is no continuous analog scale. Things are quantized. There are only certain bands at certain energy levels. There's the Valence Band. There's the Conduction Band. (And in between there's the Gap Band) That's why it's called Quantum Mechanics.
-----
In my next post I'm going to propose a recipe for digital grain. No guarantee that it will be the best. Feedback welcome.
-----
UPDATE 7/15/2009
(click to enlarge somewhat)
I got this illustration from the photo.net argument thread. I'm adding in as a partial response to the first comment below about film grain being crystals vs filaments. If I follow it, (at least some) crystals (somehow) look like filaments at the right magnification. I'm including this filament idea mainly because as of my posting of this blog item it was the strongest argument I could find for film being analog. I still think of film grain as being composed of random areas of black or clear crystals.
Comments referring to the "Gurney-Mott development model" (causing the silver speck to grow like a traditional AM halftone?) or "plate disc theory" (not seeing this on Google) are new information for me. Thank you for that. I'll investigate further.
----
THE GOAL
I should add that I don't particularly care whether film is one thing or another. My goal is to get an understanding of whatever it is - a mental model - that will help me come up with some sort of recipe for trying to copy it in something like Photoshop. And if it turns out that existing recipes for faking grain are as good as it gets, then so be it. But I want to take a closer look.
----
UPDATE
I'm looking at this page (reprinted from an old Kodak source) discussing latent images and the Gurney-Mott Theory, and it looks to be a pretty good explanation.
Here's what I read:
- The dark spots on the negative are metallic silver. They are essentially opaque. This suggests two states: dark and clear.
- The numbers of dark spots on a grain increases with exposure, but the number of grains doesn't increase. This suggests something like an FM screen.
- The dark spots tend to cluster at favored formation points on the grains. This means the dark regions grow with exposure, more like an AM halftone with jittered placement of halftone cells.
So I think after reading this material that we are talking about a 'binary' state black/white process that functions like an analog process (like a newspaper halftone), especially at the scales at which we use it.
If you are willing to call a traditional newspaper halftone an analog process, since those dots can be any size and don't take discreet size change steps, (and I see that traditional halftoning is indeed considered analog) then I guess I have changed my mind.
I still consider film to be akin to a halftone, but over the course of this posting my opinion of halftones has changed. If a black circle can be any size on a white field then I'm comfortable saying that the area of the black circle is an analog function. But is the 'gray' analog? There is no gray. But I guess the perceived gray is analog, continuously variable, too.
Okay. Fine grained film, like coarse newspaper ads, is analog.
Although this guy has taken a patent out on the notion of space-filling curves (like the filaments in the photo at the top of this blog?) as a form of digital, not analog halftone. If the coils can be arbitrarily dense then shouldn't his process be considered a form of analog halftone?
...Now how best to simulate artificial 'film' grain in a digitally-originated image?
Interesting, I don't do that much film development (these days I'm digital) but I think that film isn't binary.
During the 1970's i did some work for an image science paper for the RPS and I feel that for film to be binary it would have only 2 states obviously that's not the case .
I note that you say it resembles steel wool, and is there or isn't but surely that is due to the development process? Film grains start out as crystals and final crystal shape depends on the developer. I think the wire wool stuff is from low ph medium activity developers like MQ types, if you subject it to different developers the crystal can either stay as it was when exposed or grow in size during what's termed 'physical development' which hasn't been used for many years.
You say:
"I think after development it's little black crystals of reasonably similar size that scatter and overlap. The more light a region gets, the more numerous the crystals become."
I think you'll find film is layered in structure the crystals are there before development (during exposure) they don't really move or become 'more numerous' as such but the more light that strikes the crystal the greater the potential for reduction by the developer becomes. The crystals can have varying thickness and transmit different amounts of light- not really a definition of binary. They certainly aren't black/clear otherwise overlap would be less relevant to final image tones. Most text books I've seen suggest that they are of varying density a la Gurney-Mott development model, the old belief that film grains were black and blocked light 100 per cent is called the plate disc theory and dates back to the 19th century.
You also state:
"a Kodak scientist who flat out said film is analog, but Kodak needs to hold to the line that film is superior to digital"
I'd go with the Kodak scientist, not because he's biased but because he is probably one of the worlds foremost expert in that field- Nestor Rodriguez and whether film is analog or binary isn't really arguing that film is superior after all to most peoples minds digital is superior.
I think some of the links you gave have errors, but to me the worst piece is the bumble bee guy- he seems to have a bee in his bonnet, is is a terrible article-scientifically speaking.
I've been thinking and digging out some of my old books. The biggest problem with the argument is that grain is black or clear.
If so how does colour development fit in?
The developed silver grains are removed leaving a dye cloud, this dye cloud varies in both size and density so passes light according to the density of the cloud?
Grain can develop as a filament or not depending on the developer, but those larger grains can pass light depending on the density of the filament (no two look the same) If you look at the picture at the top of your page you have several strands how much light does it pass?
Remember also that the grains can be several layers deep which is quite unlike a half tone and furthermore the grains are coated in different speed layers (to give wider latitude) so grains are different sizes in different layers.
Also when you state:
"The dark spots tend to cluster at favored formation points on the grains. This means the dark regions grow with exposure"
Those favoured points are sensitivity specks often found at defects in the crystal, the dark points DO NOT grow during exposure but rather during reduction (development)
Good books to look at are "The Theory of the Photographic Process" C.E.K Mees
Those favoured points are sensitivity specks often found at defects in the crystal, the dark points DO NOT grow during exposure but rather during reduction (development)
Good books to look at are "The Theory of the Photographic Process" C.E.K Mees
That is my understanding based on what I've been reading, and when I say 'with' exposure I don't mean 'during' exposure, I mean in a manner 'correlated with' the exposure.
---
Grain can develop as a filament or not depending on the developer, but those larger grains can pass light depending on the density of the filament (no two look the same) If you look at the picture at the top of your page you have several strands how much light does it pass?
Remember also that the grains can be several layers deep which is quite unlike a half tone and furthermore the grains are coated in different speed layers (to give wider latitude) so grains are different sizes in different layers
If it is really a clear/white tangle (and you may be right that I am wrong to think this, but so far I still do) then that's a frequency-modulated stochastic halftone to me. Draw an arbitrary box around the overlapping filament tangles and wouldn't the gray be the perceived ratio of clear to dark?
---
I've saved the biggest challenge for last: your comment on color. This I need to check more into. You may be changing my mind here.
If the grain (on each color layer) remains dark opaque spots in a colored field, then my mental model is still ok. But I don't think that's what I'll find. The color mixing would also have to be additive to form white in an image, which, as far as I understand it, is not the case.
If the dark portions of the grain become colored elements in a clear field then the grain probably is passing along its color as a transmissive filter does, which means it takes on transparent color in different densities (grays).
I can't immediately picture a mechanism for opaque, but colored, overlapping grains in a clear field passing color along, but I want to check a little more.
Interesting.
I can't immediately picture a mechanism for opaque, but colored, overlapping grains in a clear field passing color along, but I want to check a little more.
Found it. The opaque particles are not dyed, they are replaced by dye:
Color film processing requires that the black metallic silver be removed or washed away, leaving the color dyes on the negative.
The Art and Business of Photo Editing, p. 54
by Bob Shepherd
http://bit.ly/KR8pH