When 2400 Dpi Isn't 2400 Dpi

I decided to write some software to help me measure things under the microscope, and it has been working pretty well. Yesterday I used it on a scan (Epson V300 @ 2400 DPI). I was surprised that my virtual Kiusalas gauge didn't match the perfs on the image.

So I decided to measure the DPI of a fresh scan with glass scales graduated in 1/10 of an mm and and 1/100 of a mm. Turns out my scanner is 2390 DPI horizontally and 2399 vertically -- here vertically means in the direction the scanner head moves. The measurement is very consistent between scans.

It's not much, but it's really noticeable when you are looking at a blowup of a stamp on a 30 inch monitor and the dot's don't line up.

Anyhow I thought it was an interesting observation.

Scanners can face geometric and measurment errors particularly around lens distortions and charged couple device/CCD misalignment. Also scanners perform a transformation of info from the analog to digital using coordinates based systems via complex algorithms. There are probably tradeoffs made in those ICs (chips) onboard to deliver cheap consumer scanners and probably meet overall tolerance for advertised specs.

I am not surprised by your findings in that you probably have to to spend more than $100-200 to get high precision machines gear more toward pro/prosumers class hardware.

When I did mapping software work in a previous lifetime the impact of these things proved to be more dramatic where accuracy couldnt be compromised. Jn the 1990s, high end scanner gear was super pricey.

I took a more pragmatic approach to measuring perf via the scanner. First, I scanned my perf gauge, then used the image to recreate the perf gauge's grid and scales as a png file with a transparent background. I can then move this image across a stamp's image to measure the perf. I am not at my computer station right now but I can post a picture if anyone is interested.

I actually use two perf gauge PNGs, one vertical and one horizontal so the images are always in step with potentially nominal resolutions. The results seem to be in line with published perf. When it comes the actual resolutions you mention, I'm not sure that the differences from nominal are relevant.

I observed something similar a couple of years ago with my scanner (Cannon Lide 700F). As I recall, horizontally at 1200 dpi, one inch was 1201 pixels on the left side, and 1206 pixels on the right side of the platten. This was consistent over more than one scan.



Jan

The size difference may be intentional. Prints from photocopiers were once something like 0.1 of 1% smaller than the original to be able to catch paper money counterfeiting. When you copied large format stuff, you'd notice the size difference.

https://illumin.usc.edu/making-a-qu...-in-america/
see "Counterfeiters find a new workplace", especially the last paragraph

This was my approach too when I mostly used the scanner. That's why I didn't notice the discrepancy untill I added a second imaging device to the mix.

Dimensionally, and as RIS mentions, the center of the platen is more accurate than the outer edges. The CCD is not the length of the bar (on many models it is slightly off center) so a lens (think fish eye) is used to reach the outer edges. The lens introduces distortion and this can impact the dimensionally accuracy.

But this is not an issue for perforation gaging which is intentionally 'rounded' by our hobby. Just because we now have two/three decimal point tools (perf gauges and scanners) does not mean that 'back in the day' they were dealing at that level of accuracy.
Don

This intriguing observation inspired me measure DPI variation across the platen on my brand new Epson V600 scanner. So I wasted a couple hours this evening taking several dozen 2400 DPI scans of a glass microrule placed at different locations.

Horizontal DPI varies just as you suggested. In the center 3 inches of the platen a 50mm scale measures 4744 pixels (about 2410 DPI) at every vertical location. When placed in the outer 3 inches, a 50mm scale measures 4759 pixels (about 2418 DPI) at every vertical location.

Vertical DPI on my scanner is almost precisely 2400 DPI regardless of location on the platen.

My wild guess is that vertical DPI accuracy is a function of scanner head movement accuracy while horizontal accuracy is mostly a function of optical accuracy of the scanner head itself.

The new scanner is a significant improvement over my 10 year old Epson V300! It will be interesting to redo this experiment in a few years and see if the results change.

For most practical measurements a delta of 20 DPI from 2400 DPI is irrelevant; however, it could be significant for things like like differentiating rotary vs flat printings, small vs large perforation holes, or making Kiusalas gauge measurements.

Regardless, the little OCD devil on my shoulder prompted me to make a few software changes to my image processing workflow to compensate for the delta anyhow... ;)