Film grain simulation for extreme upsizing?

[Pete Berry]

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

You mean sorta like MIT proved a few years back that the bumblebee was incapable of flight?!

Seriously, where do you see oversharpening artefact in the detail-rich areas, aside from the "grain" enhancement that the OP was looking for? There's certainly no haloing I can see in the above ~50% crop, nor unnatural local contrast enhancement, and the detail looks quite natural to me, particularly from a couple feet away - remembering that this is a representative section of a 26x40" enlargement.

Pete

[pfigen]

Well, the image you posted does have huge sharpening halos around the dust spots, and based on those, I'd have to agree that it's grossly over sharpened. On top of that the way the grain itself is rendered looks like it's been reticulated. For me, that would be an unacceptable scan.

[Pete Berry]

Well, the image you posted does have huge sharpening halos around the dust spots, and based on those, I'd have to agree that it's grossly over sharpened. On top of that the way the grain itself is rendered looks like it's been reticulated. For me, that would be an unacceptable scan.

To say a scan is grossly oversharpened purely on the basis of surface dust spot halos is really pretty silly, isn't it? The question is how the detail looks, and to my eyes it looks quite natural at this level - no halos or unnatural local contrast enhancement as would be seen with USM at this level - particularly at an appropriate viewing distance for a 40" wide print. But as they say, YMMV...It does show that grain-like structure can be pulled out of a transparency with the V-700 -argueably the best consumer flatbed film scanner - at 4800 dpi.

And yes, it would be an unacceptable scan for me also unless I was into grain showing - which can easily be modulated with the pixel radius chosen.  These scans were smart sharpened 500%/3.0 px radius - pretty much the upper limit without marked reticulation and highlight blocking as I said previously.



 

[pfigen]

Pete - If you look closely at your scan, it's pretty objectionable even in the areas that are sort of sharp. You just don't notice it as much there. You're not really gaining any detail, just local contrast. Look again at your unsharpened raw scan. It's really pretty soft even where it's sharp. What I don't know is if there's more to be had on the film with a better scanner or if that's all there is. My guess is that that scanner, when scanning at 4800 is probably pulling a good 1800-2000 if you're lucky.

What I do know is that when I scan film at 4000, I have sharply defined grain and detail without any sharpening applied in the scanning stage or in Photoshop. In fact, the only time I apply sharpening to my drum scans is when they're downsized for output. If they're output at full resolution, you typically would either want no sharpening or very very little. You might, however want to use high radius low amount for some areas, but that's a different purpose.

What kind of film was that in that scan? I must have missed that bit.

[torger]

Flatbed scans don't get that sharp, and have also problems with dynamic range. Even with precisely tuned focus using custom height-adjustable film holders an Epson V750 don't deliver more than 2300 ppi or so of real resolution of what I have seen. However, they are a lot cheaper and take less space than a drum scanner :-). They're also cheaper than medium format tabletop scanners. If one should only scan 35mm a dedicated tabletop film scanner for that is better though.

http://www.filmscanner.info/en/ has many detailed reviews and tests of scanners.

There's no doubt that drum scanning is best. My hope is that I can with the macro lens setup exceed flatbed scanners and match or exceed 3200 ppi tabletop medium format film scans and approach 4000 ppi 35mm tabletop film scans. Resolution test target transparency is on its way...

Concerning grain the idea I had was to break up aliased grain after upscaling to a finer grain structure than was sampled, since the real grain structure is indeed very fine and requires unreasonable high sampling resolution to not be aliased. Now I'm taking a step back though to verify resolution first.

[torger]

I've done some preliminary resolution testing, may make a real post with crops etc later. Anyway, using the USAF test slide I tested both 7D and 5Dmk2 at 1:1 macro distance with the latest Sigma 150mm macro lens. The lens holds up corner sharpness very well, but a little CA correction is useful.

Some numbers:

- 5Dmk2 at f/8 - ~3600 ppi, 91% of pixel resolution.
- 7D (APS-C) at f/5.6 - ~5100 ppi, 85% of pixel resolution.
- 7D at f/8 - quite noticable hurt by diffraction - ~4400 ppi, thus only 75% of pixel resolution

To get that high pixel resoultion absolutely top notch demosaicer algorithm in the raw processor is required. Only very small reduction of sharpness in this lens corners.

Compared to other tech:
- Flatbed scanners - Epson V750 Pro ~2300 ppi - only 40% of pixel resolution
- Film scanner 35mm CanoScan FS4000US - ~4000 ppi ~100% of pixel resolution
- Film scanner medium format Reflecta MF5000 - ~3050 ppi

I have a CanoScan FS4000US so I have used that as reference. I'd use the FS4000US for 35mm, or stitch/HDR with the 7D at f/5.6. I've improved my setup and technique so for the 22x15mm APS-C area the short f/5.6 DoF is manageable (one need to be extreeeemely thorough when focusing, I refocus for every repositioning, using precise leveling head and focus rail in collaboration), so I can do 5900 ppi digitalization with 5100 effective resolution. -3/0/+3 stop HDR on the 7D gives nice dynamic range, exceeding the film scanner and should be as much as one would ever need. Without HDR film scanner wins clearly, and I'd say it is to poor to use in prints. The greatest disadvantage of the FS4000US is that it is hard to get 100% of the picture area, one easily loses 0,5 mm or so top/bottom. When stitching one gets a margin.

For medium format the 5Dmk2 at f/8 plus stitching and HDR is very competitive compared to alternative tech. Flatbed scanners just suck, and the affordable and some fairly high end scanners too only do ~3000 - ~3200 ppi, sometimes with limited dynamic range. The stitched and HDR-merged result exceeds in resolution and DR. I've just shot a 60x45 medium format slide with the 7D, but that's really tedious - 16 frames need to be stitched, with the fullframe one can do it in 4, or 6 to get a bit more overlap. Most time during shooting is verifying and adjusting focus after repositioning.

The conclusion is that I think using a macro lens one can achieve better results than most think possible, but one needs a very sharp macro lens, small pixels on the sensor (to get the resolution) and very carefully executed technique. The next step up in quality seems to be a drum scanner.

[artobest]

Sounds like a pain in the butt. This is not a solution for everybody, although it may work for you.