About blooming and halation

Hi, I'm still in the process of writing an analog film emulation, and for now I have achieved satisfying results on the "basic" functions (spectral response, density function, MTF, grain, etc) - you can see some results in my previous questions.

I am now more focusing on "defects" simulation, and there are two that are linked I think, and I have an interesting question.

Blooming and halation are caused by some sort of "excess" of light. Blooming sort of "spreads" or is scattered in the emultion layers, and then for halation some light that wasn't absorbed can pass the last emulsion layer (the only one in B&W, and red one in color films), rebound at some angle and come back to the last layer, adding more brightness at other spots.

My question is : at which point does the light spread in blooming, and (I think this is linked) at which point does the light gets through the layers ?
Can it be seen on the technical data charts (I think about the density chart), or deduced? For example over some threshold in light intensity?

(I already successfully simulated these effects with more classic image processing and the results are satisfying, but now I would like to insert them in the more general process of my film emulation)

AbsurdePhoton said:

Can it be seen on the technical data charts

Click to expand...

Not AFAIK. It's also wavelength-dependent, and for any given film, it'll make a massive difference whether it has anti-halation measures built in or not. For instance, popular films like the Fomapan series have quite decent anti-halation in sheet film and 120 roll film, but lack this in 35mm format. Yet, the datasheet is the same for all products.

Btw, going out on a limb here - I'd say that 'blooming' is the visually apparent effect/outcome, and 'halation' is one of the mechanisms that can be responsible for it (imperfect lens optics being another obvious mechanism). Furthermore, I'd say 'halation' and 'light piping' are comparable effects, although the latter is more often associated with choice of base material while the former is used in conjunction with descriptions of film stack design. The way I see it, light piping in the base will contribute to halation, which will show up as blooming. You could make a conceptual model out of this, describing the different causal relationships, and use that to further explore the topic.

Quantitatively speaking, I don't think I've ever come across concrete data for practical products.

Thanks for these thoughts. Just one thing : separating "red", "blue" and "green" range photons can be a good approximation of different wavelengths. Light piping and halation seem not to be the same from what I understand.

I read that blooming mainly occurs by light spreading (diffusion) and scattering in each layer.

My thoughts. All the photons that weren't absorbed by the dyes/silver grains are the ones that can scatter/spread (some percent of initial intensity), can be absorbed again, but also can go through the layers at different angles (the rest of total intensity minus absorbed minus [scattered/spreaded and not absorbed again]).

Halation is only created by photons that get through the layers, rebound at the back of the film (much less with a anti-halation layer) and come back to the layers in reverse order (for color much for red, much less for green, negligible for the blue one, and the "only" layer for B&W). Maybe 1 to 3% of original intensity? (could it correspond to some part of the toe in the density chart?)

So for me blooming and halation are really two different things. Halation just adds more to (reinforces) part of the blooming effect, and as the photons rebounce at an angle, can possibly arrive at spots farther than the blooming effect itself, but not too far either because of the very short distances in the cross-section of the film itself.

It’s all basically the same thing, stray photons. I don’t think it particularly matters how they get where they go so long as you understand what they tend to look like in the final result.

Unless you’re making something with massive overkill like a ray tracing model.

You can emulate these effects with just a luminance key

AbsurdePhoton said:

So for me blooming and halation are really two different things.

Click to expand...

No doubt; the gist of my message was to stimulate you to conceptualize how the different effects relate. How you define each of them - well, that's debatable and some desk research should be able to clear up most of it. And in the end, it doesn't really matter how you define things, as long as you're consistent.

Here's the kind of thing I'm thinking about:


Notice how I define (my own, subjective definition!) Blooming here as an image effect. As contributing causes I discern Light piping and Halation as separate, but also causally related effects. I consider Blooming as a single construct, because in the final negative, it's often impossible to determine what the specific causes were for the non-image density to form - it all tends to show up as a halo around an area of intense illumination, and only through rather subtle and sometimes indirect clues can we figure out what may have caused it.

Now, I don't post this to say that the above is the only correct representation. The message here is to try and bring conceptual clarity to your thinking, and then use that insight to decide how you want to model this behavior in your simulation. Also note that the conceptual model above is far from complete, plus the existing parts could be refined (split up into underlying factors & relationships). How far you want to take this depends on what's needed to get the job done, and that in turn depends on the goal you've set for yourself. You'll have to make your own choices there.

You could even expand the diagram to include other aspects of image characteristics such as grain etc. - but I understand you have already tackled those, so perhaps you have little interest in going through a conceptual phase for them. It may still be valuable for the documentation of your project, though.

Just my $0.02; see if you can take anything useful from this.

Romanko said:

Point spread function - Wikipedia

en.wikipedia.org

Click to expand...

This is how I simulated bloom in a "classic" way, using an inverse Fourier transform with nice image "kernels" (which also allow to produce light flares). Thanks anyway, this was a good hint.

koraks said:

Notice how I define (my own, subjective definition!) Blooming here as an image effect

Click to expand...

Thanks for this too, I already have something like that: you can't write hi-level code without a bit of thinking and this kind of diagram.


And again thanks to all of you, you helped me think a little out of the box and I *think* I got it now.
I won't take in account lens considerations, this another module for me, and also set aside light piping for now.
Simplifying a lot, I'm actually using a source image, make heavy spectral computations using the film spectral response charts to convert pixels' intensities to exposure values. These values are then passed to the HD curve to obtain densities, then "levels of gray" for each layer, which allows to obtain both B&W and color films simulation.

Until now I was clipping some values: after a certain amount of exposure the result is for the moment some maximum density, using the HD curve (near the end of the shoulder).
I am thinking now about using the "extra" (not used) exposure to compute bloom (which will work after all with a threshold just like my "classic" implementation). The amount of exposure left after bloom computation (if any) will be fed to the halation part.
This system will allow to naturally produce strong or weak effects with the incoming exposure "extra" variations. And this way there can be blooming without halation too, which I think is how things works for real.
And it makes sense: with high exposure values, there are lots of photons, some will "miss" halide crystals/dyes (especially if the maximum density is attained) and get lost elsewhere to produce first bloom then halation. As exposure raises, "lost" photons will increase too.

AbsurdePhoton said:

at which point does the light spread in blooming, and (I think this is linked) at which point does the light gets through the layers ?

Click to expand...

I already suspected this, sort of.


If I am wrong somewhere please tell me, maybe I missed something important?

AbsurdePhoton said:

Until now I was clipping some values: after a certain amount of exposure the result is for the moment some maximum density, using the HD curve (near the end of the shoulder).
I am thinking now about using the "extra" (not used) exposure to compute bloom (which will work after all with a threshold just like my "classic" implementation). The amount of exposure left after bloom computation (if any) will be fed to the halation part.

Click to expand...

This is the part that I did not understand. The dynamic range of the camera (in stops) is approximately the same as that of the film. Why do you need to clip values?

AbsurdePhoton said:

Until now I was clipping some values: after a certain amount of exposure the result is for the moment some maximum density, using the HD curve (near the end of the shoulder).
I am thinking now about using the "extra" (not used) exposure to compute bloom (which will work after all with a threshold just like my "classic" implementation). The amount of exposure left after bloom computation (if any) will be fed to the halation part.

Click to expand...

You seem to have run into the situation, where a preliminary simulation of blooming created a nice point spread, but not the situation you see in real life on film. The likely reason these two don't line up is sharpness effects from development. Even non sharpness enhancing developers will eventually run into developable Silver Iodide, which releases a powerful restrainer.

So you don't have any real clipping effects, but a low pass point spread function (from light spread of all kinds) interacting with a high pass point spread function (from sharpness enhancement). Artificial thresholds won't address these interactions correctly. You either combine these sharpness enhancing effects (of e.g. D-76) and light spread into one module and simulate developer effects only as a delta from your reference developer, or you accept, that blooming simulation alone yields unobservable intermediate results and then add developer effects to get the real thing.

@Rudeofus I know, I set aside partially the development part, thinking that all I needed was the "light" part... but thanks to you I'm beginning to think I missed something really important.

Yesterday after reviewing all the comments here, I started to search about light scattering and found something called the Rayleigh scattering effect (and NOT the Mie theory that I read about in some academic paper that was wrongly applied to film layers). I will try this for the light spread function in layers.

Until now I was just using a one-to-one relationship: one pixel in = one pixel out. I'll have to rewrite things to create an "accumulator" of incoming light from several sources - direct, blooming, halation, maybe piping later. Add a more serious development part.

Wow, what an interesting journey. I started this as a fun side project, asking myself if I could do as well as commercial products. I rewrote my engine more than 15 times, as I was discovering I misunderstood this or that, until I got a satisfying result. Now I see that it is not enough, again. Well... (spits in his hands) there's work to do

Bonus : why Mie theory doesn't apply here - it is primarily concerned with the scattering of light by particles that are on the same order of magnitude as the wavelength of light, such as water droplets, dust, and other atmospheric particles. It explains effects like fog and haze, halos in the atmosphere for example. When it comes to silver halides in analog film, the situation is somewhat different, because the cristals are orders of size smaller: about 0.05 to 10 nanometers. That's where Rayleigh scattering applies.

It's above my pay grade to contribute to this thread, only to add that I've seen B&W pictures by a noted fashion photographer (whose name I can't remember) that seemed to employ something like halation to artistic effect. I've always wondered how that was achieved. Someone suggested some sort of infra red....

Arthurwg said:

It's above my pay grade to contribute to this thread, only to add that I've seen B&W pictures by a noted fashion photographer (whose name I can't remember) that seemed to employ something like halation to artistic effect. I've always wondered how that was achieved. Someone suggested some sort of infra red....

Click to expand...

There was an IR film by Efke with no anti-halation-layer: Efke IR820 Aura ... plenty of opportunity for creative stuff as you can imagine.

AbsurdePhoton said:

@Rudeofus I know, I set aside partially the development part, thinking that all I needed was the "light" part... but thanks to you I'm beginning to think I missed something really important.

Click to expand...

So far you were able to break this into a linear part and a pixel local non-linear part, which made simulation doable. Things break apart, when this non-linear part radiates back into the so far linear part. That's when all these "easy" simulations break down, since you can no longer model them as consecutive convolutions in 2D space.

You will likely reach this point as you start merging sharpness effects into your simulation, and my previous statements imply, that you have to do this very soon now, since without this your simulated blooming will not look right. This was the point when my own attempts to simulate this went awry, so I have no idea how to progress from there :-(

My point of view is, the simulation of an analog process using digital means doesn't have to be 100% correct if the result looks faithful. There's always a way to make it better, after (do you know about the wheel of improvement?). That's what I am experiencing with my desire to make my simulation better. At first sight, it looks silly to add defects, no? But finally it showed the limits of what I had achieved so far.

For example, I'm at the point where the simulation is already good as it is, and is at least as good as several commercial products. I could compare several scenes taken both with digital and analog cameras (and several types of films). I applied my method to the digital pictures, and the results were comparable to the analog pictures.

Here is an example found here.



On the left the digital version, center the Ilford Ortho+ version of the same scene, and on the right my Ilford Ortho+ simulation. The digital version has much less details in reds than the analog version and impacts the simulation, but if you look carefully: the gray levels look good, as well as their distribution. And this was just from a little JPEG file, without color calibration!

With several examples like that, with different films, I knew I was on the right path... and it's not finished.

Another example: from the same website, I used the test target:



Ilford Ortho+ on the left the analog version, on the right my simulation (without grain).



Now Ilford HP5+ (with grain)



This one looks a bit different: grays and blue and yellow look good, green and red are too light - maybe more control on the development would help