About resolution of film scanners
by Lorenzo Comolli
First written in 1998, last update in 2006.
Thanks to the anonymous readers who reviewed my original English text, and corrected its grammatical errors. Any residual error is due only to the author and it's updates.

With advances in computer technology it is possible to digitally enhance our astronomical images. The process of digitization of the images becomes a problem to be seriously considered. If we want to get the maximum information from our images, it is much better to scan the original negative or slide than to scan a print, which already has lost some original detail. Commercial film scanners have resolutions of 2700 dpi (in the price range of under $1000) and 4000 dpi (<$2000) and from 24x36mm film (35mm format) thay produce files of about 50 and 130 MB (at 16bit/channel, RGB). Those scanners are considered to produce high resolution scans and with good reason. But when we look at high resolution astrophotography with hypersensitized Kodak Technical Pan (unfortunately discontinued in 2005), a resolution of 2700 dpi is found to be inadequate. Only the problems associated with scanning Tech Pan will be examined in this article.

The film

Tecnical Pan is a very high resolution film. Measurements show the grain size to be 3 micrometers. The smallest star sizes that can be achieved with this film are about 9-10 micrometers. The pixel size of a 2700 dpi film scanner is also 9-10 micrometers. This is not sufficient for resolving the smallest stars as one pixel would equal one star and the result would be a square star. It's assumed we need to oversample the star by a factor of two or three to achieve adequate resolution. This implies an optical resolution of 5000 to 8500 dpi. (These numbers are not exact multiples of 2700dpi but are the resolutions I have chosen to use for the graphical simulations on this page.) Ideally we would like to resolve the grain of the film but that would require 25,000 dpi and would generate a 5 GB file from 35mm film! This is beyond the capability of most computers in use today.

The simulation (1)

The image above shows what we lose when using a scanner. (Click the image for a higher resolution version.) A Tecnical Pan negative is shown at full resolution in box (1); note that it's not a real image but a digital simulation. The stars in the upper row of each box are 20 micrometers and the stars in the lower row are 10 micrometers. Scanner resolutions of 8500, 5000 and 2500 dpi are shown in boxes (2), (3) and (4) respectively. It is easy to see that a scanner resolution of 2500 dpi is completely inadequate to resolve the stars. A scan at 5000 dpi is a good compromise between resolution and file size. At 8500 dpi the grain of the film begins to resolve. Even a scan resolution of 8500 dpi cannot be compared to a traditional print, but there are many advantages to be had by digitizing the image and enhancing it with software like Photoshop.

 The simulation (2)


Click on the image for the hi-res version, because this little reproduction does not give a correct idea.
 This simulation is surely more realistic than the preceding because it's based on a VERY HIGH enlargement print from a Technical Pan negative realized from one of the best Italian astrophotographers: Aldo Radrizzani (unfortunately passed away in July 2005). The telescope is a Schmidt camera with 30cm free aperture at f/2: this instrument gives very little stars, at the limit of film capabilities.
My idea was to realize a very high enlargement print, to digitalize it and to degrade it to the resolution of various scanners (2700 and 4000 dpi) and to an ideal resolution (8000 dpi). In this way it's evident what we lose in digitalizing a negative. Note that the results are in agreement with the simulation (1): 2700 dpi are not enough for Tecnical Pan, 4000dpi are a little better, but 8000 are much better.

Now a consideration of dimensions: Aldo's negative is a dish of 77x77mm diameter. How large will the files be for a 77x77mm image at the various resolution?

77x77mm film
 Scan dimensions [MB]
Resolution
B/W, 16bit/channel
RGB, 16bit/channel
2400 dpi
 106
 318
2700 dpi
 134
 402
4000 dpi
 294
 882
4800 dpi
 423
 1270
8000 dpi
 1176
 3529
9600 dpi
 1694
 5082

With such dimensions it's obvious that it's hard to work on the images. With actual PCs (January 2006) we can work on images of max 300-400MB, and also on those we risk to use swap files and virtual memory that slow down the work considerably. We have to wait many years, maybe 5, before PCs with the needed power to work on 8000dpi images will be available at reasonable prices.

The scansions

After the first draft of this article, now are avaiable good film scanner at prices under 500$. For Aldo's films I needed a large format scanner (to scan 100x120mm film). Up to now there are no dedicated film scanner, at reasonable prices. So I've selected a flatbad scanner with film support, with a good resolution (4800dpi) and high quality (16 bit/channel and low noise). My choice is the Canon CanoScan 9950F that can scan film up to 200x300mm!
So I've started to scan and elaborate the Aldo's films, the BW ones at 4800dpi  (file of more than 400MB) and the color ones at 2400dpi (more than 300MB). The elaborations needed a lot of time on my PC with a AMD Athlon64 3200+ with 1GB of RAM (that will be 3GB in short...). On the average I need an hour to scan and elaborate at best a BW negative. The work is hard but the results are impressive, it's incredible to examinate and navigate in images of 15000x15000 pixels.
The upper simulation revealed to be confirmed: at 4800dpi the stars are well resolved and also some grain is visible. And so this is the better  compromize for scanning the while archive of Aldo. The results are very good, so that they can be passed to next generations on digital support, preserving the work from the unavoidable degrade of the negatives.


Scansions with CanoScan 9950F: comparison between 2400dpi and 4800dpi. Zoom at 400% and 200% respectively. This is a field north of M31. Please note that the grain is better resolved at 4800dpi, even if not completely. IMHO, the little quality increase justify the more used memory.


An example of one of Aldo's round negatives: the North America nebula complex and Pelican. Click on the image for hires.

Conclusions

With actual instruments the scansion of Technical Pan start to be possible, even if to obtain the best results it is necessary to enlarge the image by 3 or 4 with a slide duplicator, prior to scanning. Scanning the 35mm film directly results in the loss of the smaller and fainter stars by about one magnitude. This loss is caused by the pattern of pixels that sees the film: imagine that you have a star (of the same dimensions of a pixel) with a S/N (signal to noise ratio) of 3 and that you scan it as a square exactly centered on the star. You'll have a digital S/N of 3; but if the star is centered on the corners of four pixels, than the S/N becomes 1.75 (=1+(3/4)). So there is a loss.
A frequent error that is made in evaluating the adequacy of a film scanner is to underestimate the importance of film grain. Experience shows that unresolved grain translates into a smoothing of the background sky thus making even a professional digital print lights years from a traditional print.

Thanks to Alfredo Zanazzo for the long discussion about this problem and to Aldo Radrizzani for his huge film archive.

Send me your comments or critiques on the ideas expressed here, e-mail at: comolli@libero.it. (Send me also any corrections for my not very good English! ;-)

HTML Editing and Publishing by Lorenzo Comolli. Email me at comolli@libero.it.
Back to Main Page