Scanning old photographs question

Discussion in 'Beginner Questions' started by tomazdrnovsek, Jun 28, 2020.

  1. Epson claim a lot of things - like 3200 dpi (sic) 'true optical resolution' - that are pure hyperbole.
    The Nyquist limit only applies to the maximum frequency that can be captured before aliasing occurs. This is a spatial frequency in the case of a scanner, so 3200 samples per inch can theoretically capture 1600 line-pairs-per-inch. This is about 63 lppmm, and far more detail than could reasonably be expected in any 10"x8" print.

    WRT distortion: Not a practical issue IME. You can scan graph-paper with most scanners and get a very close match to the original spacing.

    However, I'd still question the claim of 3200 ppi across the whole of an 8.5" platen. When I cared about such things I scoured the data sheets of all the tri-linear CCD sensor manufacturers (Toshiba, Sharp, Dalsa, Kodak, etc.), and could find nothing that offered more than 10,200 pixels across its length. This would give a 'resolution' of 1200 ppi over 8.5".
  2. It is hard to find so much detail, but it does seem that Epson uses 6 line CCDs.
    It might be that three rows are offset by half, such that it can get the higher resolution.

    I haven't looked for the data sheets on them.

    As for distortion, I suspect that it isn't really an issue, but it does complicate the meaning
    of resolution if you use the correct meanings of accuracy and precision.
    You can have one without the other.

    If I have a tape measure with 1mm divisions, I expect to be able to measure things with
    accuracy close to 1mm. Sometimes I don't need so much accuracy, but only need

    Practices of Science: Precision vs. Accuracy |

    Just saying resolution doesn't say whether it is accurate and/or precise to that value.
  3. I totally agree.

    From browsing Epson's website, it appears that you're quoting the specification for their Perfection V100 scanner. This scanner sold for about £70 UK, and if the specification is to be believed, it potentially out-resolves their V800 Pro model at ten times the price. I find that highly unlikely and difficult to believe.
  4. I didn't look for the V800, but have the service manual for the V700.

    (Which doesn't say near as much as it should.)

    The V700 says "4800dpi optical resolution". (yes, it says dpi) at 8.5in for 40800 pixels,
    or 6400dpi times 5.9in for 37760 pixels.

    (Maybe the manuals are made by the same people that make the printer manuals.)

    It seems that it switches lenses between the two modes. It does say it is a six line sensor,
    but doesn't give details on those lines. My guess is six lines of 20400 pixels spaced such
    that it can scan like three lines of 40800 pixels. (Reminds me of how printers work.)

    The one I have is the model 3200, where it seems to be numbered after its "3200dpi"
    (quoting Epson again). I believe it is older than the V series, but might have been at
    a similar place in the product line, at the time.

    For many scanners, the back light for transparency scanning is not the full width that
    it uses for reflection scanning. That might be where the 59 inch comes from.
  5. Since none of Epson's scanners achieve anything close to their claimed "4800 dpi true optical resolution", or whatever figure; I think the mechanics of how they fail to achieve it is purely academic.
  6. The V700 has the ILX185, but I haven't found the data sheet for it.
  7. It is hard to find data sheets, but I found the ILX137K:

    ILX137K (SONY) PDF技术资料下载 ILX137K 供应信息 IC Datasheet 数据表 (1/13 页)

    Which seems to be designed for (so it says) 2400dpi scanners.

    As above, it has six rows, two of each color offset by half a pixel,
    of 10700 pixels each. There are microlenses, but it doesn't say
    what it does with them.

    Many of the web sites are in Chinese, making it harder to read them, though
    the actual data sheet is in English.
  8. Thanks for that link Glen.
    As I suspected, the horizontal pixel number is just over 10,000; making the PPI figure over 8.5" about 1200. Not too bad for a relatively cheap scanner, and total overkill for most reflective copying purposes.

    The six (2 by 3) rows of photosites appear to have the purpose of speeding up the scan time, by increasing the distance between scanning steps. There's no way it can improve resolution in the direction of scan that I can see. Which must remain at 1200 ppi due to the line spacing.

    So, once again Epson are found out in their distortion of the truth.
    Microlenses are there to increase light efficiency - so called 'fill factor'. That's always their sole purpose.
    Last edited: Jul 27, 2020
  9. The rows are one half pixel offset, to get in between. That one claims 2400dpi
    I believe that the elements are. close to the full width, so a funny spatial frequency response dip,
    but you already said that doesn't bother you. I am not sure which scanner uses that one.

    I haven't found the data sheet for the ILX185K, though, but I believe that one has twice as many for 4800 across 8.5in.
  10. They're one half photosite offset, not half a pixel. A pixel only exists after the photosite data are processed.

    So from the photosite geometry shown, it would be possible to extrapolate 2400 pixels per inch, but only in one axis. The vertical axis can only support 1200 ppi because the photosites are asymmetrical and rectangular in shape and spacing. Therefore the vertical pixel value must be interpolated between two offset rows of photosites.

    The overall pixel accuracy also depends on near absolute mechanical placement of the subject relative to the sensor, or vice-versa. Requiring a stepping accuracy of less than 2 microns x whatever the magnification ratio is.

    All pixelated images have a 'funny' spatial frequency response. The pixels' spacing varies depending on their angle relative to orthogonal. At 45 degrees, for example, the pixel spacing is 1.4 x that at 0 and 90 degrees.

    I never said that the spatial frequency response 'doesn't bother me'. But as long as it's high enough, and at a sufficient amplitude for the intended purpose of the image, then why worry about it?
    Last edited: Jul 27, 2020
  11. They could make it more obvious.
    The whole array is 42.8mm long, with 10700 pixels (what they say) with 4um pitch.
    I thought it was 4um along the array axis, but am not so sure now.
    OK, so 2.4um wide cells on 4um centers.

    Then there is the cylindrical lens, I believe over each of the six rows.
    I think that narrows the effective height, but it doesn't explain that part.

    Scanners are supposed to scan with increments smaller than the cell size, though with plastic
    bands that might stretch moving the mirrors around.

    I used to read comp.dsp where audio signal processing was a common discussion topic.
    Frequency response is very important in audio.

    Give the MTF graphs for film, it does seem that photography isn't so interested in it.
    The response of the wider cells might not be so far from the MTF graphs that
    go down to 10% or so.
  12. Ah, OK.
    That's a different kettle of bananas.
    Audio sampling is essentially one-dimensional and ideally at a single point, or over a very small time-period, per sample.

    Whereas image spatial sampling is 2 dimensional by necessity, both in sample area and as a re-assembled data matrix.

    Ideally those sample areas would tesselate with no space in between, and it's this that the microlenses attempt to simulate. The actual photosites need space in between for electrical interconnects.
  13. I still have a Nikon Coolscan V dedicated 35mm negative/slide scanner, but it only still works because of Vuescan since Nikon abandoned their software. You would think that even with film as a niche product that Nikon could make a steady profit by providing quality scanning of 35mm. Apparently not. I don't use film much these days but it's nice to have as long as it lasts. Most old pictures that I have scanned in I've used my Epson Perfection V500 because what negatives I had were large (and it was mostly prints). Epson did a good job with that scanning. I scanned all my Dad's family pictures to show to my relatives. Alas I only got them when my dad died so I couldn't really ask him about them.

Share This Page