A couple posts back I compared 35mm slide scans made with my Nikon Coolscan film scanner and my Canon 9000F flat bed scanner. I mentioned that my normal workflow was 3000ppi scans on the Canon and I concluded that the Canon optics weren’t good enough really compete with the Nikon. But I still thought the Canon scans were just fine for my purposes. After reading Steinhoff’s book I learned some things that seemed to mostly confirm my experience, but also made me wonder if the user (me :() might be part of the problem.
In The VueScan Bible Steinhoff makes these points that prompted me to do another experiment:
- “A scanner can scan properly only at its optical resolution (e.g., 4800 spi) or parts of it (e.g., 2400, 1200, 600, or 300 spi).”
- He mentions that a scanner’s effective optical resolution can be dramatically lower than the advertised capability and he specifically mentions the Canon8800F (a direct, close ancestor to mine) saying “The Canon CanoScan 8800F has a nominal optical resolution of 4800 spi. The effective optical resolution is roughly 1600 spi–quite a big difference.”
- Using a Nikon similar to mine as an example, he also writes “…for maximum image quality, you should use output size reduction… Configure Input > Scan resolution a 4000 spi and Output > size reduction at factor 2. This will convert the 4000 spi scan into a 2000spi file-effectively, the file size will shrink from 160 MB to 40 MB! But the main reason for using output size reduction instead of scanning directly at 2000spi reducing image noise. Shrinking a 4000spi scan to a 2000spi output file has a positive effect on image noise–there is simply less of it left. That’s the reason why can be justified to scan 4000 spi, even when your desired output size is only 2000spi.”
Interesting points, and if he’s right on all counts, some of my choices might be contributing to the issues I’ve seen and there might be hope to get better results from my scanners. So, I devised an experiment to see if I could find out whether it mattered or not.
Bottom line is that it does seem to matter and I’m adjusting what I do because of it.
I picked 3 slides and scanned each in both scanners at the scanner’s optical resolution, saving them at full resolution and with output size reduction factor 2 (4000ppi-to-2000ppi or 4800-to-2400ppi). I also repeated my scans at 3000ppi which isn’t a match for optical resolution on either scanner. In all cases the sharpen filter was ‘off’ and the scratch/dust filter set to ‘low.’ The results confirmed his points:
1. The difference between the 3000ppi scans and the optical resolution scans was subtle but real. The difference was much more pronounced with the Canon-both the scan saved at 4800ppi and the one at 2400ppi were better than 3000ppi (ouch!).
2. This experiment and all my experience with the flat bed is 100% consistent with what he says–limitations imposed by the Canon’s optics made them less sharp than the Nikon. But…
3. In all three cases, the difference between full resolution saves and the size reduced saves was negligible unless viewed full, pixel-for-pixel on the screen.
In fact, for two of the three slides the best looking scans were Canon scans at 4800ppi saved at 2400ppi. These were a portrait (samples below) and a landscape. Only with a high contrast, close up of a flower, did the 4000ppi Nikon scan look better. Yes, the Nikon scans were consistently sharper, but for normal viewing or printing, the Canon output colors had an edge and the Canon’s infrared dust/scratch removal seemed a bit better.
I thought about posting all the slide scans but doing the test created over 2o scans, some are really big files and some of the differences were difficult to see in the post. Instead I decided I post a few examples so you can get a feel for just what the Canon is, and is not, capable of.
This pair is the crucial result, the Canon scan on top and the Nikon scan below. In both cases the scan was done using the VueScan software, at optical resolution with a down-sample of x2, scanned 4800, saved at 2400 in the case of the Canon
and 4000, saved at 2000 in the case of the Nikon. This pair can be examined full size and I think you’d agree that there are subtle differences but I found both of them to be nice.
Indeed, the differences here are interesting and consistent with what I saw before and during the experiment. The Nikon’s scans are sharper (look at the eyelashes), slightly less saturated in color and a bit cooler in color. Which one is right? I don’t know–looking at the slide with a loupe on the slide sorter I’d say neither is what I see there but both are close. Neither example has had post-scan editing, but with just a little tweaking either could look like the other or how ever I decided was “right.”
Although the scratch/dust processing was on in both examples the experiment included on and off. The two scanners use different technologies for dust/scratch removal and each has its advantages. The VueScan software knows how to get the best out of the Canon’s hardware in that regard and I liked those results better.
Down-sampling pros and cons
Purists may object to the idea of the x2 down-sampling approach and with some justification–why throw all those pixels away? I thought that way, too, until I’d read Steinhoff’s book and checked him against another author as well. The truth is that you’re not merely throwing them away but using the information in all the pixels to construct a better, and smaller image. The experiment confirmed what they were saying which is that yes, you’re giving up a something in one way, but you’re getting some processing gain which results in less noise in the output–better images. And the reality is that with the Canon I’m losing nothing since its optics can’t actually do 4800ppi anyway. The image above is the full resolution version of the Nikon scan (all 14MB of it) for comparison with the down-sampled version. The other benefit is smaller file size:
|Canon 4800-to-2400ppi||3.3MB JPG|
|Canon 4800ppi||12MB JPG|
|Canon 4800ppi||26.5 MB TIFF|
The JPEG files were saved with the minimum compression setting (highest quality) in VueScan. The results were essentially identical for the Nikon scans: the down-sampled files were just about 25% as big and the TIFFs were 2-4x the size of the higher resolution JPEG file.
Just in terms of effective resolution, the Canon or, I suspect, any flat current bed can’t compete. If I really have to have biggest, sharpest scan I can get, the Nikon at 4000ppi is the way to go. They’re sharper, but they’re more work and give big(!) image files, especially if saved as TIFFs. But I really like the results I get from the Canon now.
So, I expect that for >95% of my future slide scans, the Canon scanned at 4800ppi and saved at 2400ppi will be my choice. That seems to be the best trade-off of time invested, file size and quality of output. I still wish the flat bed scanner’s optics were better, but by changing the workflow a bit, I think I’ve gotten to the point where I’m getting the most out of the Canon that I can and I’m pretty happy with the results.
If anyone else has similar (or contrary!) experience with film scanning I’d love to hear about it, too.