Film grain simulation for extreme upsizing?

[torger]

I'm shooting transparencies with a macro lens. Matching/outresolving actual detail in film is no problem, 4000 ppi seems reasonable for that, but to get grain you don't seem to get enough resolution.

I've attached enlarged crops of shots taken with a 150mm macro lens at 1:1 with a 5Dmk2 (~6 um pixels =>~4000 ppi) and 7D (~4 um pixels =>~6000 ppi) and 7D with 2x teleconverter (~2 um pixels). A bit softened by diffraction from f/8, and enlarged. That grain detail is there at 12000 ppi should be evident though. Disregard from color/contrast differences, has not been tuned.

Say if sampling at 4000 ppi, yielding ~20 megapixels for a 24x36mm slide, prints do not look pleasing at close range if print resolution gets lower than ~200 ppi, grain becomes unnaturally blotchy and you start to see pixel patterns (I hate pixelation, but kind of like the look of grossly enlarged film).

Digitizing at 12000 ppi just to get more grain detail becomes awfully unpractical though, 180 megapixels for just a 24x36mm slide! So what I'm wondering is if there is any upsizing software that does film grain simulation, that can make that 4000 ppi digitalization look more like the 12000 ppi digitalization - the difference is mostly random grain, and that the shapes of high contrast edges does not follow a pixel pattern so I was thinking that could be created artificially.

[Schewe]

...I'm wondering is if there is any upsizing software that does film grain simulation, that can make that 4000 ppi digitalization look more like the 12000 ppi digitalization - the difference is mostly random grain, and that the shapes of high contrast edges does not follow a pixel pattern so I was thinking that could be created artificially.

PhotoKit Sharpener 2 has a variety of film grain routines...

Alternatively you can roll you own. Add a layer filled with 50% gray set to an overlay blend mode. Run a couple of Add Noise runs at different %'s (what % you'll need is resolution dependent). After the noise run g-blur a time or 2 with a radius of .3 or .4. You may want to come back and add a gentle touch of USM after the blur....vary the opacity to taste.

[bill t.]

FYI there is an upsizing program called PhotoZoom Pro 4 that has a "Film Grain" slider.  It is a very basic grain pattern and won't look anywhere near as filmic as Schewe's algorithm.

And there is quite a bit of irony here for this guy who worked so hard during his analog youth to smack down film grain!

[Schewe]

And there is quite a bit of irony here for this guy who worked so hard during his analog youth to smack down film grain!

I love film grain when it's useful...

If you are shooting 4x5 or 8x10, you ain't looking for grain, you're looking for grainless.

But small format film had a grain that added a patina to the resulting printed image...grain ain't "bad" unless it interferes with the look of the final printed image.

[Bart_van_der_Wolf]

FYI there is an upsizing program called PhotoZoom Pro 4 that has a "Film Grain" slider.

That's a program I was about to suggest. And unless you really want to add fake film grain, adding a little noise to a huge enlargement will help to mask the featureless magnification quite effectively. Remember, human vision doesn't look for film-like grain, it looks for a suggestion of detail. Noise does that better than film grain.

One could also let the noise have a Poisson distribution (e.g. by using ImageJ with the RandomJ plugin, Poisson/modulatory), just like photon shot noise, and let it's amplitude vary with the average brightness. Or add a noise with a spectrum to only affect the highest spatial frequencies (with ImageJ one can create a FFT bandpass filter that only passes the highest spatial frequencies of a uniform noise). For calculation efficiency one can create a decently sized patch that can be seamlessly tiled to fill any size output at the final output size.

Cheers,
Bart

[deanwork]

The more recent version of Blow-Up is outstanding for those situations where you have to upsample a file for output. It also has a feature that allows you to add and control simulated grain that works very well for me.    http://www.alienskin.com/

There is also Real Grain by the same company that makes Noiseware, http://www.imagenomic.com/rg.aspx

You can test both demos. I have Blow Up and use it for grain but if I had to do a lot of this I might try out Real Grain as well. I it apparently has more capability but I haven't tried it.

john

[pfigen]

Torger - Looking at your "scans", as they are, you're not resolving anywhere near the grain, and what you're seeing certainly is not very sharp. I think you need to see a real 4000 dpi and 8000 dpi drum scan for comparison, scanned with a 6.35  and 3.17 micron spot, and you'll really see what I mean. I think whatever your setup is, is not nearly as sharp as you think it is. Send me a piece of film and I'll rip off a couple of reference scans and send 'em back for you to see. What you really see is that very few films show any real benefit beyond 4000 dpi, but a few like Kodachrome 25, late K64, T-Max 100 and of course the old Tech Pan, as long as they're shot with sharp lenses, will benefit from the higher res.

[torger]

Torger - Looking at your "scans", as they are, you're not resolving anywhere near the grain, and what you're seeing certainly is not very sharp. I think you need to see a real 4000 dpi and 8000 dpi drum scan for comparison, scanned with a 6.35  and 3.17 micron spot, and you'll really see what I mean. I think whatever your setup is, is not nearly as sharp as you think it is. Send me a piece of film and I'll rip off a couple of reference scans and send 'em back for you to see. What you really see is that very few films show any real benefit beyond 4000 dpi, but a few like Kodachrome 25, late K64, T-Max 100 and of course the old Tech Pan, as long as they're shot with sharp lenses, will benefit from the higher res.

The photo is of Fujifilm velvia. The lenses used for the original shot was probably not very sharp, an indication of that is that there is quite a deal of chromatic aberration towards the edges of the film. Not sure how well-focused the shot is etc either. It is not my own photos, but a photographer I know which did not use very expensive equipment. I have compared the macro shots with a CanoScan FS4000US 135 film scanner, and the sharpness compares well to that. I would be interesting to see a real drum scan though, which should exceed the film scanner's performance. The photos are also upscaled and no deconvolution sharpening has been applied so they look softer than they are. The 2x TC on 7D should however surely not be super sharp optics and be the softest per pixel, and the 5Dmk2 sharpest per pixel.

Sharp or not, one can see that the general film grain structure is significantly smaller than a pixel at 4000 ppi, so even with perfect sharpness the grain would not be represented that well. Perfect sharpness also typically means aliased, which means that they work poorly when upscaled or printed at low PPIs.

Next step for me will be to look on the film under a microscope to learn more about what it really looks like. I don't have the film archive where I live though, so it is just now and then I get to do this.

I think I shall buy a resolution test target transparency too to truly verify actual resolution of the macro photo method. It would be valuable for me if I can use it, which means I can digitize not only 24x36 but also medium format etc (requires stitching of course) at will per need basis, but if sharpness is just too bad as you suspect I would have to do drum scans I guess, but these are quite hard to come by here in Sweden, so I rather would like to be able to copy myself.

[Bart_van_der_Wolf]

The photo is of Fujifilm velvia. The lenses used for the original shot was probably not very sharp, an indication of that is that there is quite a deal of chromatic aberration towards the edges of the film. Not sure how well-focused the shot is etc either. It is not my own photos, but a photographer I know which did not use very expensive equipment. I have compared the macro shots with a CanoScan FS4000US 135 film scanner, and the sharpness compares well to that. I would be interesting to see a real drum scan though, which should exceed the film scanner's performance. The photos are also upscaled and no deconvolution sharpening has been applied so they look softer than they are. The 2x TC on 7D should however surely not be super sharp optics and be the softest per pixel, and the 5Dmk2 sharpest per pixel.

Hi,

It should be possible to get quite good results and, as you've shown, more than 4000PPI is required to resolve the graininess (grain or dye clouds suspended in 3D).

However, one important thing is possibly overlooked. The presumably diffuse back-lighting of the film has a graininess (and dust / scratches) suppressing effect. Sharpness is only affected due to smoother contrast, but also delivers a better starting point for sharpening.

The main issues with shooting film in a macro setup are avoiding light on the front of the film surface because it reduces contrast, and absolute plan parallel orientation of sensor and film. Of course at these magnification factors the lens quality can help and improve the result even further. I view the diffuse backlighting as a benefit because it reduces artifacts and saves time in postprocessing.

Sharp or not, one can see that the general film grain structure is significantly smaller than a pixel at 4000 ppi, so even with perfect sharpness the grain would not be represented that well. Perfect sharpness also typically means aliased, which means that they work poorly when upscaled or printed at low PPIs.

Scanning with resolutions above 6000 PPI (118 lp/mm) will not deliver much in the sense of resolution (film MTF usually doesn't exceed 85 cycles/mm and lens MTF added will reduce that further), but it might reduce graininess just a little more by reducing grain-aliasing. Diffuse backlighting also reduces the graininess and grain-aliasing due to lower contrast, but subsequent sharpening could strengthen graininess a bit again. An edge mask will help to mitigate the adverse effects of sharpening.

Cheers,
Bart

[torger]

As a guy interested in optimizing stuff to the best, this is truly interesting to work with :-), and I'm clearly just in the beginning of the learning phase. I'm suspecting that the 3Dness of the film and light scattering inside the film itself reduces the ability to get things super sharp too. I'll order resolution test target transparency right away though so I don't need to guess what actual resolution of this system is... I think it is better than flatbeds and slightly worse than film scanners (not a lot worse though), but a test target could give me some hard numbers :-).

[Bart_van_der_Wolf]

As a guy interested in optimizing stuff to the best, this is truly interesting to work with :-), and I'm clearly just in the beginning of the learning phase. I'm suspecting that the 3Dness of the film and light scattering inside the film itself reduces the ability to get things super sharp too. I'll order resolution test target transparency right away though so I don't need to guess what actual resolution of this system is... I think it is better than flatbeds and slightly worse than film scanners (not a lot worse though), but a test target could give me some hard numbers :-).

Hi,

If you want to save some money, you can print your own target for free. You can use the target (created by me) discussed here, and also compare to the results posted so far. It would be interesting to add scans (or macro's) of film as well.

Cheers,
Bart

[torger]

If you want to save some money, you can print your own target for free. You can use the target (created by me) discussed here, and also compare to the results posted so far. It would be interesting to add scans (or macro's) of film as well.

Thanks, I've used that test target to test some of my lenses. Problem in this case is that I need a small transparency with very high resolution rather than a paper print so I can verify the lens at 1:1 distance. The test target transparency I found at scandig.com is not too expensive though, just €59.

I just started to read about wet mounting... uhh... it's becoming more and more complex these things.

[pfigen]

I posted this but it never showed up. Thanks to the History function in the browser it was still there, or here, now:


One thing I can tell you is that even on an extremely good drum scanner, scanning at 8000 ppi, using a 3.17 micron aperture spot, the scanner will not record everything that is on the film. I had always been told that it would, but taking an extremely sharp piece of Velvia 35mm film that had been shot at f/4 with a Canon 200mm 1.8 on a heavy tripod with mirror lockup, etc. I was able to see more detail on the film than I could record with the scanner. I didn't have a microscope, but I did have a Beseler 45M racked all the way up, focusing through an Omega grain focuser, and through that, there was clearly more on the film. Not a lot, but enough to see. A microscope would have probably seen even a greater difference.

The problem, of course, is how to get it off the film, as every optical system has its losses. In my experience, the least amount of optical losses are in the drum scanner's optical system, which only has to record an area far less than a square millimeter at a time, using the sharpest part of a very expensive optimized lens versus, in this case, a macro lens recording the entire image at once. The drum scanning/digital print combination seems to be the best way to get the sharpest prints with the best tonal qualities compared to any other, including direct optical printing, which seems to incur more overall losses with the combination of enlarging lens and paper resolution than the scanner route.

I don't have the reference on this computer and don't feel like researching it right now, but there are people out there obsessed with this sort of stuff, and have measured on medium format cameras, with the Mamiya 7 80mm lens being one of the best overall, actual on T-Max100 film resolution numbers of 120 lines per millimeter in the center of the lens. I think the surprising thing also was that some of those medium format lenses far outperformed the best 35mm lenses, challenging the common wisdom that the smaller the format, the sharper the lens. If we only had Technical Pan still readily available.

[torger]

I found this to be an interesting read for a beginner like me: http://cool.conservation-us.org/coolaic/sg/emg/library/pdf/vitale/2009-10-vitale-filmgrain_resolution_v24.pdf

It has some numbers on film resolution.

[Slobodan Blagojevic]

http://goo.gl/wdCrh

I'm shooting transparencies with a macro lens....

There is an interesting post by Ctein, over at the The Online Photographer, about that.

[Pete Berry]

I'm shooting transparencies with a macro lens. Matching/outresolving actual detail in film is no problem, 4000 ppi seems reasonable for that, but to get grain you don't seem to get enough resolution.

There appears to be little sharpening in your examples, so I gave a try at rather radical sharpening to see what detail/?grain I could pull out from some recent scan slides on my V700 @ 4800 dpi.. First, with Smart (deconvolution) sharpening, I pushed it to 500/3.0/lens blur removal mode, after seeing "grain" detail start to appear at pixel radius of about 1.5. I then sharpened a bit further with USM at 50/1.0/0 to further define it. The amount of detail enhancement and it's quite natural look with Smart Sharpen was pretty surprising to me - something well beyond what USM alone could do in my experience.

Is the effect in the 100% crops attached anywhere near what you're looking for?
 
Pete

[torger]

http://goo.gl/wdCrh
There is an interesting post by Ctein, over at the The Online Photographer, about that.

Ctein kind of reminds me of Ken Rockwell, I really can't handle that style of writing - i e when all things are claimed to be simple absolute truths, when there in reality is a lot of buts and nuances. When I'm myself experienced in the subjects written about I can recognize when the line of bullshit is approached or passed, but in this subject I can only recognize the writing style and be warned. I would suspect that the statement that a 24 megapixel camera can only represent half that worth of resolution is a plain factual error (you really don't lose that much from AA and demosaicing), but perhaps all macro lenses really suck at 1:1 and that what was he meant, I don't know. I've ordered a resolution test target transparency so I'm about to find out resolving power at 1:1 of my lens at least.

What I have from my own work so far noted is that HDR is necessary to get sufficent dynamic range, and to keep the sampling resolution one need to stitch larger images. To shoot a 24x36mm slide at 1:1 I needed 3 x 4 = 12 APS-C shots for HDR and covering the full area.

Should all this end up being as bad as a flatbed scan, it is not worth it. If it is almost as good as a 4000 ppi 35mm film scanner and a little bit better than a 3200 ppi medium format scanner, then it is worth it. Managing the grain will be necessary for any of those methods though.

[torger]

Is the effect in the 100% crops attached anywhere near what you're looking for?

I've studied film scanning more and more for each post in this thread, so I'm not really sure what I want any longer :-). This is breaking new ground for me. What looked simple to start with, no longer seems...

Initially my plan was to scan and archive at a reasonable resolution, but still being able to make rather large prints (I like that) at high quality without needed to specifically rescan for that. I'm not really afraid of that the "lofi" quality of 35mm and budget lenses shines through (it is almost a plus with true analog lofi when it is nearly extinct), what I'd like to avoid is any signs of insufficient quality in the digital printmaking workflow. Doing say a 40x26 inch (100x67 cm) enlargement of a 36x24mm will bring out that lofi look of the original, but since a 4000 ppi digitalization then becomes 140 ppi, one starts to see traces of digital up close.

Grain aliasing giving unnaturally coarse grain and that one starts to be able to see that the picture is builtup from a digital grid of pixels are my two concerns. So concerning grain what actually is needed is to break up and create finer grain on the uprezzed version than there is in the scan. I've got a few tips of things to try out in this thread, I haven't had time to go through it all yet though.

Now my primary concern is to verify the actual effective resolution in this digitalization procedure. I'm not going to use this workflow if a flatbed scanner is producing equal or better results. I do have a dedicated 35mm film scanner (not running exactly now though due to a driver problem), but no MF scanner and there are some medium format transparencies I want to digitize too, that's why I started off with this method. I'm kind of verifying it on 35mm first though.

[Pete Berry]

I've studied film scanning more and more for each post in this thread, so I'm not really sure what I want any longer :-). This is breaking new ground for me. What looked simple to start with, no longer seems...

Initially my plan was to scan and archive at a reasonable resolution, but still being able to make rather large prints (I like that) at high quality without needed to specifically rescan for that. I'm not really afraid of that the "lofi" quality of 35mm and budget lenses shines through (it is almost a plus with true analog lofi when it is nearly extinct), what I'd like to avoid is any signs of insufficient quality in the digital printmaking workflow. Doing say a 40x26 inch (100x67 cm) enlargement of a 36x24mm will bring out that lofi look of the original, but since a 4000 ppi digitalization then becomes 140 ppi, one starts to see traces of digital up close.

Here's what the original looks like scaled to your 26x40" size - a little better than 50% original, and close to 180 ppi from the 4800 dpi flatbed V700 scan. The grain prominence is easily manipulated up or down through the deconvolution sharpening settings.

Pete

[deanwork]

That's over sharpened because the original is simply not sharp enough to make it sharp using any software in the world.