[jonby]

This is interesting. Nice post.

The 5D can see 21x10^6 pixels. This works out as 150 pixels per mm, which in turn is 75lp/mm. I doubt the Epson can see more than about 25lp/mm. Sharpening etc doesn't add anything to the resolution, it just increases the apparent contrast of low frequency details.

Decent film, shot through a sharp normal, like the 80mm (I've got a Bronica and saw a test on line for PE lenses, Hasselblad and Mamiya, and I seem to remember the Mamiya glass was highly rated) will reach 100lp/mm. There are Zeiss docs online recording 400lp/mm, but this is unlikely unless you have awesome film and glass, shot with specific lighting conditions etc. But 100lp/mm is feasible.

The fact that neither the 5D nor the Epson hit 100lp/mm means that they need to look at an enlarged section of the negative and then have it stitched together. The Epson cannot do this.

If you look at this link:
http://www.addicted2light.com/2012/11/23/best-film-scanner-canon-5d-mark-ii-...vs-epson-v700/

its clear that the stitched 5D scans exceed the Epson by a long way.

Some interesting observations and info here. I agree with much of what you say. The tests I have shown suggest that with this particular negative, a small amount of very fine detail is not being captured at 1:1 magnification with the 5D II, and therefore a slightly higher magnification would need to be used in order to capture this, requiring the stitching of more shots to get a complete frame. Finer grained film like Ilford Delta would certainly need a significantly higher magnification to capture all available detail.

The examples and comparisons shown on the link you give are, based on my own observations, misleading however. Information given about how the examples were produced is quite sketchy and in some cases, contradictory, making it difficult to draw reliable conclusions from them. For example in the first comparison - the black and white crop with the rocks, the shooting magnification of the camera shots or scanning resolution are not given.

Based on my own experience, I would say that the 5D image was shot at a magnification approaching 1:1 which would require several stitched shots to cover the whole frame. Also the scan does not appear to match the quality I get from my V700. As can be seen from my example given above, the scanner is capable of capturing a reasonable impression of the film grain, whereas here the grain is completely smeared away. So I think the examples on the linked page overplay the difference between the two methods substantially. Of course it's possible to get far higher quality by using a camera at very high magnification than a consumer flatbed scanner. But the higher the magnification, the more shots have to be produced, manipulated and stitched, which is a very lengthy process if you want to do a good job. Comparing like with like (ie. a single scan versus a single camera shot), the V700 (or at least my copy) certainly out-resolves the 5D II when copying a 645 frame.

---------- Post added 02-26-14 at 07:40 PM ----------

I just had to try this, but I don't have a light box, or a negative carrier, or even a flash. So I set up my tripod in front of the desk, put a flat white image on the pc monitor and cranked up the brightness, and suspended the film between The Lord of the Rings and The Silmarillion using rubber bands. This ghetto setup worked surprisingly well, and in many ways the result compares favourably with my developer's scans. Oh, and I managed to enhance the negative with a big fat greasy fingerprint after the experiment...

From top to bottom:
1. Whole frame from DSLR
2. 100% crop of scan
3. DSLR crop scaled to match scan
4. 100% DSLR crop

The scan shows obvious macroblocking; I don't know if they are compression artifacts (1.5 MB JPEG) or some scanner idiosyncracy. The DLSR version is very soft, but compared to the scan it looks much more pleasant up close. I may have missed focus, something may have moved during the 1-second exposure, and/or the focal plane may not have been parallel to the film plane - I suspect all three.

The film is Ilford Pan F Plus exposed in an Olympus OM-10. According to the EXIF data, the scanner is a Noritsu QSS-32 or QSS-33 (listed as "NORITSU KOKI QSS-32_33"). Raw file from Pentax K-5 with SMC D-FA 100mm WR. Coarse lighting adjustments and monochrome conversion in PDCU. Inverted, cropped, and compressed in GIMP.

Jens Lyn, this is a good effort considering the equipment you had to work with. Your shot is clearly a lot better than the commercial scan. It might look softer but there's far more detail from the film in it. I'm surprised you managed to capture so much of the grain with a K5 and a fine-grained film like Pan F. Did you shoot the whole film frame or just a portion at the maximum magnification of the lens (1:1)?

---

Last edited by jonby; 02-26-2014 at 07:41 PM.

[whojammyflip]

Yes, I think that to get the most out of a scan requires some skill, by the sounds of it, however, the flatbed scanners are limited to 30lp/mm, which means if you want the type of scan a stitched photo can produce from a DSLR, your only option is to go to a drum scan. I had been interested in the sound of the Plustek 120 scanner, but 1) its expensive 2) by the sounds of it, its temperamental. All things considered, if a photo is really worth printing, its probably easiest to spend an hour in the darkroom on it! I am only really interested in the scanning from the perspective of building a journal of my photos, to work out which ones I need to print. Really, what I should start doing, is contacting printing each roll, which would have the added benefit of telling me whether the negs are incorrectly exposed (too thin or dense in the low zones) and for establishing standardised development and darkroom exposure times. There is a good book by Carson Graves which covers the value of contact printing, which I nearly sold on Amazon and then held onto when I realised the value of this information. Anyway, this is getting into the realms of APUG!

Another idea, off the top of my head, would be to enlarge using an enlarger, onto a piece of ground glass, lying on top of a flat bed scanner. Could this work? Then you would be getting the benefit of top optics in the enlargement stage, so that the scanner would be looking at an image with maybe only 10lp/mm of detail, but spread over an A4 sized area.

---------- Post added 27th Feb 2014 at 09:53 ----------

However, you can see from his histograms how dark and compressed the red channel is compared to the green and especially the blue channel in the original image following inversion. In order to get 'normal' colour balance and contrast, the red channel must be stretched and brightened to a high degree. The same information is being stretched over a wider brightness range, which reduces quality.

this is something which gets missed a lot...the bit depth...something which is obvious in a black and white wet print

[Jens Lyn IV]

Did you shoot the whole film frame or just a portion at the maximum magnification of the lens (1:1)?

I shot the whole frame. If I try this again I might stitch 2 frames shot in protrait orientation, APS-C being roughly half the size of 24x36 film. I might also try a shot with the holes visible so I can use them as a reference for sharpness.

[whojammyflip]

I shot the whole frame. If I try this again I might stitch 2 frames shot in protrait orientation, APS-C being roughly half the size of 24x36 film. I might also try a shot with the holes visible so I can use them as a reference for sharpness.

I have to stress that I dont think its worth anyone's time to go mad like the bloke I'd linked to, with masses of photos being stitched together...the major advantage of this process is that its quick, significantly faster than the scanning I did with a v300.

[Jens Lyn IV]

He he, curiosity trumps convenience...

[jonby]

I shot the whole frame. If I try this again I might stitch 2 frames shot in protrait orientation, APS-C being roughly half the size of 24x36 film.

Yes I think you should try this (so that I can see the results!). I've always thought that a two-shot stitch might be worth the effort but have never got round to doing it.

There are a number of difficulties in getting images which can be stitched seamlessly without much effort: exposure variation, lens vignetting, keeping the shots parrallel, lens distortion and aberrations are a few of the minor ones. Those associated with film flatness are the hardest to overcome, however, as this causes small but complex distortions in the image geometry as well as shifts in sharpness as the film moves in and out of the focal plane.

I was a little disappointed with Ricoh-Pentax's new film duplicator accessory, which seems to offer little more than a bit of convenience over the kind of setup you have used. They could have incorporated a shifting film holder which allows for accurate multi-shot work, for example. Also they should have used a ring mount which goes between the camera and the lens. This would potentially mean that once set up correctly, there is no need to align the camera to the film plane each time you used it - one of the main headaches in this process. The ring mount would also act as an extension tube giving you greater magnification with whatever lens you are using. Sure, these additions, plus the heavier-duty precision engineering that would be required would make it more expensive, but given the high cost of medium format film scanners, it could come out much cheaper than this.

On a slight tangent, has anyone thought about super-resolution software like PhotoAcute for film copy work? I haven't tried it yet but in theory you can just about double your resolution and eliminate noise using several shots with tiny differences in position. This could theoretically work with scanning too.

[Jens Lyn IV]

several shots with tiny differences in position

Would this be an application of interferometry? I thought that was reserved for megabudget astronomy...

This sounds like the kind of feature that could be implemented with a firmware update. Depending on the amount of computation required, the whole process might be possible in-camera.

[jonby]

Would this be an application of interferometry? I thought that was reserved for megabudget astronomy...

This sounds like the kind of feature that could be implemented with a firmware update. Depending on the amount of computation required, the whole process might be possible in-camera.

I don't know much about interferometry but I think there might be some relation in that they're both to do with combining multiple measurements/samples in order to get more information about the subject.

Super-resolution involves taking multiple exposures with sub-pixel-sized shifts in position of the image. Analysis of these images makes it possible to work out the nature of details smaller than a single pixel.

Yes in theory this could be done in-camera in much the same way as in-camera HDR works. I can only think of three reasons why this feature hasn't appeared in cameras so far - 1) The required processing may be too heavy to be practical with current processors, 2) licensing the technology may be too expensive, 3) lack of imagination!

This does seem like a no-brainer for Pentax if it were possible. You could more or less double resolution and massively decrease noise. For tripod work Pentax has the advantage that they could use the sensor shift to take a series of images which is optimal to get best results (ie. by shifting the sensor by sub-pixel amounts over a series of six exposures, for example).

[whojammyflip]

interferometry involves taking two rays of light and superimposing them, simultaneously, when they're exactly in phase, they add together, and when out of phase, they cancel out...this is wavelength dependent, so that some wavelengths cancel and others don't due to differences in multiples of wavelength involved...examples would be the mutli coating reflection having a colour to it, or coloured bubbles

alternatively, you can measure distances using a laser which is being modulated at a known frequency (so that the carrier wave has a much longer wavelength than the actual light) etc

there is one example of superposition of waves where the waves don't have to occur at the same time...this is when a single photon of light can be shot out through young's slits and you still get double slit diffraction patterns which would occur from the interference of two beams of light, but we know there is only one photon which has been sent, so it should result in single slit patterns....an argument for parallel universes

I don't think Pentax cameras will be doing this any time soon...

---------- Post added 3rd Mar 2014 at 11:02 ----------

Double-slit experiment - Wikipedia, the free encyclopedia

[Jens Lyn IV]

He he he.

"Full frame? 36 MP? That's cute - my interferometric quantum mechanical Pentax camera with WPDMOS* sensor can shoot parallel universes!"
*Wave-Particle Duality Metal-Oxide Semiconductor
Batteries and sanity not included. Carefully read instructions (sold separately) before use.
This infomercial may or may not be endorsed by Ricoh-Schrödinger Imaging.

I think 'oversampling' is the word I was looking for...