Pimp my scan: How to increase resolution by software!

Discussion in 'Film and Processing' started by christian_bartling, Dec 6, 2012.

  1. Pimp my Scanner: Increasing flatbedscanner-resolution by software!
    Hello everybody,
    I am from Germany (so please indulge me for some false terms) and after some years of going digital have taken the fresh challenge of analogue photography, for reasons of limited space in my home with a hybrid workflow.
    I´d like to describe my way of increasing the limited resolution of my Canoscan 9000f, which is not 9600 ppi but just about 1700 dpi (according the testresult of www.filmscanner.info ). I scan 35mm-negatives and slides with the pacific imaging proscan 7200 (in Germany distributed by the brand Reflecta) with decent results, the nominal resolution of 3600 ppi results in optical resolution of 3250 ppi (same source), wich is quite nice.
    MF-material is being scanned in the above mentioned Canon-fb-scanner, wich means, that almost all of the resolution-advantage of medium format is lost due to limited scanner resolution.
    As not everybody is owning a Nikon-highend-scanner, the following may be of some interest to others:
    The software photoacute3 ( www.photoacute.com) was developped to increase the picture quality of mobile phones and digital cameras. It does so by merging multiple pictures (with minimal shift from one picture to another) which results in higher resolution with less noise. What I did was scanning 1 slide ten times, moving the slide a very tiny little bit with every scan. I used Vuescan, 3600 dpi, then resizing "bicubic sharper" to 1800 ppi (3.7 MP) in photoshop elements in order to get the most out of the scanner resolution. Then this 10 almost identical jpgs were processed by photoacute to a 13.7 MP picture.
    One disclaimer: This is only about image resolution, not about dynamic range and colour-reproduction. I know about the importance of 48-bit-tiff or .dng for this aspects, but this was not the topic of this trial, so for reasons of processing-speed and conveniance the scanning and processing was done 24-bit/jpg-based.
    This is the basic photo:[​IMG]
    In the following 100%-Screenshots you can see: to the left the scan done by the Reflecta proscan7200, in the middle the result achieved by Photoacute3, and to the right Vuescan and PSE (4-times multiple scan, 7200 ppi downsized in 2 steps"bicubic sharper" to the same size, sharpening as much as possible) as good as it gets. http://farm9.staticflickr.com/8487/8...a6082f3d_h.jpg http://farm9.staticflickr.com/8479/8...4c63d9f3_h.jpg http://farm9.staticflickr.com/8345/8...478c1e10_h.jpg http://farm9.staticflickr.com/8342/8...84bd7a56_h.jpg
    Allthough the performance of the filmscanner is not exactly matched, this cumbersome workflow can provide a considerable improvement when using flatbed-scanners, leading to Scans from 6x6 Slides of about 60MP (so a powerful computersystem is needed). To be honest, it is not always that easy and successful, some attempts went out like this:[​IMG]
    I guess the program is having difficulties with pictures that contain repetetive patterns (here the brickwall). This surely has to be improved to get the optimum out of it, but for the first try I found the results to be quite impressive. Has anyone heard of this (or similar) workflow before?
  2. Part 2:
    shows you the answer to my question, how many scans are necessary to get the best result.
    The Answer is: 2!
    Upper left the Input (6x6 negative, scanned 48bit, 3600 dpi, resized to 1800 directly by vuescan (no extrastep via pse this time).
    Upper right result of merging 2 pictures, lower left 4 pics. lower right 6 pics. I see no difference between 2,4 and 6 source-photos.
    This makes the whole thing a lot easier and faster. I think the highlights show better details than the source as well.
    The resulting .dng has 60 megapixels (340 MBytes).

    If anyone here owns a USAF-Resolution-Chart it would be really interesting to see what the optical resolution actually is.

  3. I find this a bit baffling. Can you post a before (9000F) and after (9000F through this software) for a clear comparison? The 4x through Vuescan looks terrible, because the scanner can't align itself very precisely.
  4. Thank you for your comment.
    Please take a look at the pictureset in the second posting (the rope). The picture upper left is a crop of a single scan (no multiexposure, no multi-pass) which was one of the source pictures for merging.
    But I will of course reexamine this to find out, if multipass worsens resolution (i haven´t observed this, so far).
  5. Now I´ve compared singlepass vs. quatruplepass in the Canoscan 900f. I see no difference, no increase and no decrease of resolution. As there is no problem with scannernoise in this picture, I see no difference regarding noise as well. So I will go on with singlepass.
    Following uploads show 100%-Crops (or at least what flickr lets you show nearest 100%) of the 3600dpi singlepass (left) vs. The merged photoacute-Result (same as shown before):
    And once more, same pictures, both resized (bicubic sharper) to 1/4 of the pixelcount (resp. 1800ppi) to see the result near the native optical resolution of the canoscan. I still like the pimped photo more.
    My personal conclusion: The "terrible" result in my first example was partly due to overpostprocessing, doing less would have been better. I apologize for that.
    But the resolution advantage is still there, even visible after resizing to the scanners native resolution.
    Good night and good luck
  6. Interesting. This batch of scans shows a clear improvement in detail.
    Does this process leave any artifacts? What about point light sources or areas of very high contrast?

Share This Page