Optimal Scan Resolution

[frank_gary]

<p> I’m a little confused as to if the higher resolutions my scanner can produce are doing any good or just adding to file size. To try this out I started running 48 bit (16 per channel) scans from 600 to 4800 DPI. I then evaluated the same physical area of the film to the same size on screen, not 100% for each scan but rather so that the “P” filled my viewing area. My thought was that once I stopped obtaining additional improvements in the area I would call that my top scan resolution and not bother losing space and scan time going higher. From some reviews I read I expected to top out around 2400 DPI, this also seemed consistent with the many “MP of film” posts I have found here and elsewhere. However, to my eye it seems that there is more information to be obtained in the “smoothness” of the letter curve and this is where I was mostly finding differences. I do feel that the cement texture is mostly captured and don’t believe there is anymore to pull out there. To confirm my feeling about more smoothness to be had I checked the image out under a 100X Lupe (read very nice microscope) and would say that what I’m seeing on screen is not everything that I’m seeing in the slide. <br>

Tech Details: Canon 900F scanner, shot on a Nikon F5 with 35mm lens, Astia Film, Vuescan for software, the images are straight out of the scanner, saved as .dng imported into elements (no changes in camera RAW) cropped down to the P and saved as full qual TIFFs so I wouldn't lose bit depth.<br>

What do you think? Is there anything extra in these higher resolutions or am I seeing nothing more than an upressing program would give me? Where would you call the upper resolution limit of this scanner? Personally I doubt that this little scanner is giving me a 121MP images (ok technically it is but you know what I mean). </p>

[frank_gary]

<p>2400 DPI only the P from the word Presto</p>

[frank_gary]

<p>4800 DPI</p>

[frank_gary]

<p>9600 DPI</p>

[frank_gary]

<p>If it would be better for me to post in a form different from TIFF please let me know I just figured that was the best way to give you the full image that I'm seeing without anything adjusting it. </p>

[don_essedi]

<p>If you want a non-lossy format for the web, browsers will display png files, but not tiffs.</p>

 

[Ed_Ingold]

<p>JPEGs are fine for posting on the web. They are lossy, but only if you save and resave the files.</p>

<p>Because of the way they work, the practical resolution of consumer-grade flatbed scanners is only about 1/2 the "optical" value, based on the sensor array. Simple lenses and robotics result in a lot of overlap between pixels. You will not see any improvement in image quality at higher resolutions, while the file size increases exponentially.</p>

[Dave Luttmann]

<p>The maximum resolution for the scanner is around 2000ppi to 2200ppi. Scanning at higher rates won't give you much....other than smoothing off some of the rounding errors. For your scanner, I'd scann at 3200 ppi and not bother with anything higher. Oh...and scan in 16 bit!</p>

[marc_bergman1]

<p>Here is a test of the Canon 9000F. This may be an updated model. It shows an effective resolution of 1700ppi. They don't write what resolution setting they used but scans at 4800ppi and 9600ppi showed the same resolution.</p>

<p>http://www.filmscanner.info/en/CanonCanoScan9000F.html</p>

<p>I have an Epson scanner. It actually shows a slight improvement with a scan setting of 4800ppi. However that probably only means it has an effective resolution of 1600ppi versus 1500ppi at the 2400ppi setting. Using this higher setting makes little difference in prints. It does make a difference in scan speed and file size.</p>

[dave_redmann]

<p>Look at it from the other perspective: 2400 ppi is 47 lp/mm. Fuji's data sheet on Astia 100F (AF3-149E, p. 8) shows Astia's response down to 50% (a commonly-used cut-off for resolution) at about 35 lp/mm, and at 47 lp/mm, the curve is down to about 40%. Obviously the subject detail's contrast can affect what the film records. But once you combine the effects of Astia's falling response with the lens's response etc., any detail finer than a <em>true</em> 2400 ppi is probably recorded on the film with fairly low contrast, and won't contribute a whole lot to the final image.</p>

<p>Now if you shoot with high-end primes, T-Max 100, a sturdy tripod, etc., you'd expect a different result. But you'd probably need a much better scanner to achieve it.</p>

<p> </p>

[frank_gary]

<p>Ok so as I understand it what is happening here is as follows. If I had three pixels next to each other one white then one gray then one black scanning at a higher resolution say 5 pixels would simply result in two more shades of gray being "interpreted" as being between the white and black pixels regardless of if there were or not. is that right. And if so does that mean that scanning at at a higher resolution is really just adding as much detail as running something like genuine fractals on the image? That link mentions reducing file size without losing the real info in the scan how is that achieved?</p>

[robert lee]

<blockquote>

<p>... does that mean that scanning at at a higher resolution is really just adding as much detail as running something like genuine fractals on the image?</p>

</blockquote>

<p>Genuine Fractals does fractal interpolation (up-sampling) plus some proprietary hocus-pocus. Personally, I think it more snake oil than anything generally competitive. In any case, if image detail isn't recorded in the first place, it can't truly be added back during reproduction.</p>

<p>Scanning at higher resolution will get at true detail on the film, assuming again that there's actually additional detail recorded on the media to be had. The huge caveat is that the scanner (the ensemble of optics, electronics, and mechanicals) has to be sufficiently capable. What your link says is that although the 900F can indeed mechanically step the CCD array at 1/9600in increments, the optics it mounts is no good for more than 1700dpi.</p>

[Dave Luttmann]

<p>Robert, I agree on the Genuine Fractals. A simple experiment to try is to take an image, upsample 200% with Genuine Fractals and also with Bicubic Sharper. Print the results. You'll never touch GF again!</p>

[robert lee]

<p>Look for a resampler, e.g., ImageMagick, that implements Lanczos resampling. It's slower than bicubic, but gives much better results.</p>

[andylynn]

<p>Flatbed scanners lie. Sure, they can generate an image file with a number of pixels that represents 9000 PPI. But given the optics and sensor and lack of focusing, the amount of detail is rarely more than 2000 PPI.</p>

[frank_gary]

<p>Ok this makes sense. Basically the lens over the CCD can't resolve more than 1700 DPI. For the sake of my understanding lets say that is 5 "lines" wide (9600/1700). When set to 9600 DPI I step one line (9600th of an inch) at a time and get the average of the 5 lines around where it should be rather than the one line I would get if everything were perfect. Thus the "smoother" color changes but no additional detail. After all I know there is detail not being pulled out as I can't see the grain as I can on my microscope :).</p>

[robert lee]

<p>That's about right.</p>

<blockquote>

<p>After all I know there is detail not being pulled out as I can't see the grain as I can on my microscope :).</p>

</blockquote>

<p>So you've already got a key component of a DIY film scanner :-)</p>