Dimensional Scanner Distortion

[chusteczka]

I have an Epson Perfection V200 Photo Scanner and an Epson Perfection V300 Photo Scanner that are both distorting the dimensions of the images that are scanned.

A perfectly handdrawn circle with a high quality compass will be scanned normally top-bottom but will be extended or stretched 1-2mm left-right.

The image has been scanned with 1200 pixel resolution in both .TIF and .PDF format.

Does anyone have experience with such distortion issues with scanners?
Does anyone have a solution?
Maybe someone knows of a scanner that does not distort the images in this manner?
Any suggestions?

The URL to the image is here:
http://pics.bbzzdd.com/users/chusteczka/scanner_distortion.png

EDIT:
Image added:

The image was drawn on paper and scanned, then manipulated with GIMP.

 

[chusteczka]

Has anyone seen this issue before? I cannot be the first.

Does anyone have a recommendation for a scanner that scans perfectly, without any distortion?

 

[gsaldivar]

If you haven't done so already, use the measurement tools built into the software (ruler, etc). They will help determine if the distortion you are seeing is actual (scanner issue) or perceived (display/resolution issue).

Try scanning the same image at different resolutions. It is possible that the scanner is doing some post processing, such as interpolation, may be introducing the distortion Good luck!

 

[chusteczka]

If you haven't done so already, use the measurement tools built into the software (ruler, etc). They will help determine if the distortion you are seeing is actual (scanner issue) or perceived (display/resolution issue).

Try scanning the same image at different resolutions. It is possible that the scanner is doing some post processing, such as interpolation, may be introducing the distortion Good luck!

Thank you for your reply. You make some good points. The Epson software possesses only scanning functionality, with no post-scan image processing abilities. At least that I am aware of.

I did use GIMP's software measurement tool to measure the distances of the image, then printed the image on paper and measured again. GIMP's measurement tool accurately measured the distortion. The first thing I checked was whether it was the printer or not, so used two separate printers and both printed the same.

Regarding the interpolation issue, that was something I looked into yesterday. My scanners both have optical scan (without interpolation?) at 4800x4800 pixels, while I am scanning at 1200x1200 since that is the capacity of the laser printer I am using. It could still be an interpolation issue, I am not sure.

I am out of ideas and just dealing with the problems for now.

Unfortunately, my partner is not familiar with creating modern computer CAD drawings and he relies on his abilities to draw by hand. His work is perfect, like art but I must scan each page into the computer and this is introducing errors into his otherwise perfect hand drawn drawings. For the computer system to introduce errors is like sacrilege to the effort he took to create the drawings.

Am I missing anything obvious here?

 

[gsaldivar]

It is often a bit difficult to determine what the true optical resolution of a scanner is vs. additional "enhanced" resolutions. Interpolation might not always be apparent as it can be done in software, or in hardware depending on the scanner. Complicating the matter is the fact that many scanners vary their engine speed to achieve certain resolutions. Many manufacturers consider resolutions achieved by altering motor speed to be "optical" as well. Here is a hypothetical example for a given scanner:

A. 300x300 = downsampling at 2x scan speed
B. 600x600 = true optical
C. 600x1200 = optical at 1/2 scan speed
D. 1200x1200 = interpolation upsampling at 1/2 scan speed

For this example, B might provide the most accurate possible optical scan, and all other resolutions provided through hardware/software enhancement.

I would make a list of all possible scan resolutions available on your scanner, then scan the same drawing at each resolution and check each scan for geometrical accuracy. If it is not possible to get one resolution that is completely accurate, then perform a different test:

Scan a full sheet of fine graph paper at your preferred resolution several times and carefully check the scanned files to see if the distortion is consistent between all the scans:

If the drawing is scaled by the same constant factor on each scan (i.e. scale is off by exactly 1.5mm every time), then you can compensate in Photoshop/GIMP by simply performing a scale on every scan to adjust for that.

All of these tests should be done without print, as you are first testing only the scanner for accuracy. After you have found a way to achieve geometrical accuracy with the scans, you can then perform similar tests with your printer (print at different resolutions, measure each for accuracy, etc.)

An additional thing to check for, this one might be a long shot: Flatbed scanners rely on very accurate motors to achieve accurate scans. As I mentioned above, the speed of these motors are often intentionally altered to achieve different resolutions. The speed of the motor can also be affected by slight changes in the power that is supplied to the scanner. Power "sags" or "dips" caused by poor electrical connection or an overloaded circuit (like when someone is running a washing machine or dishwasher on the same circuit) can potentially cause the scanner motor to run at (for example) 98% speed instead of 100% speed. This may introduce scaling problems similar to the ones you are having.

To test for this, try putting the scanner on an isolated, clean electrical circuit or consider purchasing a high quality UPS to compensate for the line conditions.

Hope this helps.

 

[chusteczka]

Thank you very much for the information. I will definitely try this.

 

[Mark R]

Is the distortion of constant scale in the horizontal direction? I would guess it should be. The scanners probably use a horizontal CCD, where each pixel in the CCD gets converted into a pixel in the image (obviously, not taking into account any interpolation).

If you can measure the horizontal scale factor, could you just simply adjust the scale in an imaging program to bring it back to square?

I.e. if a 100mm circle is rendered as a 100x102 mm ellipse, why not apply a scaling: 100% in Y direction, 98% in X direction? This would correct for any systematic scaling error.

 

[chusteczka]

I have not seen a constant scale adjustment in the software and am finding the variation is not constant.

My findings are currently inconclusive.

I disposed of the original circle and moved on with work, then received your messages providing some direction. The help from both of you motivated me to try again and this time I drew a square. With a square, scanned at the following resolutions, there was no distortion in the scanned image. I tested at:

600x600 dpi
800x00
1200x1200

These all showed the same measurements for a square at 100mm and 150mm length, leading me to confusion. In writing this, the scanner may have an issue with scanning a circle, so I will test again with a circle.

 

[gsaldivar]

These all showed the same measurements for a square at 100mm and 150mm length, leading me to confusion. In writing this, the scanner may have an issue with scanning a circle, so I will test again with a circle.

For the exact resolutions that yield accurate scans of the square, try each of those resolutions again with the circle.

Keep us posted, I'm curious to see if we can get it resolved!

 

[gsaldivar]

It occurs to me is that the scanner might be performing some other kind of automatic crop/rotate manipulation based on the perceived image perimeter. That manipulation might be flawed in a way that causes some sort of scaling error when applied to a circle rather than a square.

I might test for this by preparing three drawings and testing to see which are scanned accurately:

1. 110mm square only
2. 100mm circle only
3. 110mm square, with a 100mm circle contained within it

Good luck.

 

[chusteczka]

Interesting statement. I will try this but give me a few days. We had a minor tornado go through here two days ago and power has been out for a couple days. There is work to do but I will get back to you.

 

[grepnold]

Coming late to this discussion. I have a synchronization problem which I think is also due to this problem.

The scanner builds up images by assembling scanned lines as it moves along the vertical direction. This is done using a stepper motor and I would not expect it to be the source of this sort of distortion.

Each horizontal line is projected into the sensor array using a lens, and this would seem more likely to be the culprit. The array could be too near or far from the focus or even not be parallel to the field, which would give a scaling that varied across the line.

Lenses can also cause radial (pincushion or barrel) distortions, where the scale in the centre of the field is different from that at the edges. A lens designed for reprographics shouldn't do though.

It would help me in my own investigation if the OP or other readers would care to send me test scans by PM. I'll do the same and will report back what I find. What would be useful would be a scan of an accurate metric scale (e.g. a ruler with a centimeter scale), placed vertically and centrally for one scan, and horizontally for a second one. Moderate scale of 100 dpi is fine.

I only see this reported with the otherwise excellent Epson Perfection models (which I too an using). I wonder if it doesn't happen with other scanners.

 

[corkyg]

I suspect the problem is in the scanware. I am using an old Epson Perfection 2450 Photo flatbed, but with no specific scanware other than the TWAIN driver executed by Irfanview. The only circle I had was a round cork coaster. The result is a perfect circle with no distortion. A key setting is to preserve the aspect ratio.

The noted dimensions are of the white background with is the result of a quick manual drag and drop prior to scan.

 

[dbcooper1]

Is this fed through the scanner or static on flatbed? Can you rotate 90 degrees and see if the distortion rotates with it? Most distortion I've seen has been related to rollers and how positively they drive the paper. Recent drive mechanisms for read head CCDs are driven by stepper motors and shouldn't be influenced by minor power variations.

Not sure if there are auto feeders available for those scanners or not but it might also be worthwhile to boot a Linux live CD and scan in that environment and compare results.

 

[corkyg]

OK - here's an old TV Test Pattern - has circles. I downloaded it then printed it and then scanned it with the Epson. No apparent distortion that I can see. Same scan process - Itfanview Acquire via TWAIN.

 

[grepnold]

Thanks for replies, This is a flatbed. The distortion is not so great that is obvious to the naked eye.

I scanned a measuring tape horizontally and vertically. Superimposing the vertical image on itself displaced vertically shows no distortion - the grids line up. But copying an end segment of the horizontal scan into the centre shows that the central area is larger than the edges. I don't think there's any difference in scale between one end and the other, so if you flipped a copy of the horizontal scan over it would line up with the original. So the distortion is symmetrical about the vertical axis.

When I get a better scan example I'll post it.

 

[grepnold]

Here's an example of the distortion. I scanned a measuring tape horizontally and pasted a portion from the center against a portion from the edge.

The top one is the edge and the bottom one the center. The 2 and the upside-down 6 line up, but 40mm to the left the 9 and the upside-down 9 are a millimeter apart. So, the center is displayed 2.5% wider than the edges and is characteristic of lens distortion.

I did the same with the tape vertically, running along the axis that the scanner assembly travels. This did not show any distortion of this kind. That was as I expected. The stepper motor and mechanical coupling might give individual pixel-to-pixel inaccuracies, but not cumulative ones over several rotations of the motor.

 

[Josh T.]

I only see this reported with the otherwise excellent Epson Perfection models (which I too an using). I wonder if it doesn't happen with other scanners.

Are you still interested in researching this issue?

I just purchased a Canoscan 9000F MkII a couple weeks ago and started doing a few experiments last week before I start scanning a bunch of old photos. After scanning the same images vertically and horizontally and flipping back and forth in a viewer, it appears that I am experiencing the same type of lens distortion that you're talking about. At first I thought it was just a scaling or mechanical precision issue for X vs. Y, but after doing a couple of 1200dpi scans of inch machinist rulers and measuring the pixel distances of each inch, it appears that the X-direction pixel value of an inch varies from 1194 to 1204 in a six-inch span (+/- 0.33% variation from nominal 1200dpi). That's not very much, I realize, but it's a bit disconcerting seeing the differences when flipping between scans of the same 4x6 photo, so I'm in the process of at least trying to understand the phenomenon (even if it's inevitable) before I commit the time to scanning all my old photos, and this thread is the only information I've found online that mentions flatbed scanner lens distortion.

 

[Josh T.]

I did some further experimentation with this scanner at 600dpi, and it appears that the distortion is progressive (as one would expect from lens distortion). Interestingly enough, the distortion appears non-centered on the platen and does not follow a symmetrical pattern across the width.

[Edit: removed image because it posted full-width. To see image, go to: http://meatballrocketry.com/misc/RulerCompare.jpg]

[Edit: I just added half-inch pixel measurements to the ruler scan, as well, and when looking at it at that scale, the asymmetry seems less drastic. So it appears that scanning images at the center is still the best option to minimize distortion.]

 

[grepnold]

Hi Josh. My interest is mainly for a task that involves automatic synchronisation on a pattern, and I don't know yet whether the degree of distortion will be a problem for that. I was wondering whether it was just a problem with Epson Perfection models, as both the OP and myself found it with that scanner, and there's little if any other mention of it online.

You're finding it in another make of scanner, though at levels getting on for an order of magnitude smaller than I did. I'd hazard a guess that this level of distortion is common across the range of scanners. I believe lenses exist that are true to scale, but no doubt there are tradeoffs needed in practice.

I can easily imagine the distortion can be asymmetrical. All it would need is for the assembly to be a degree or two out of the horizontal. I'd have thought that the lens distortion itself would be symmetrical and being out of parallel would give predictable results, i.e. I wouldn't expect anything that couldn't be modeled with the right simple equation.

I can see that it would be quite possible to correct the distortion using a corresponding opposite horizontal distortion. I'm sure a processing tool like ImageMagick could approximate it, or a sample scan could generate an exact transform.

For my own task, since your model scanner is so much better, I'll just try to find a scanner that is good enough for what I want. My Epson, though getting old, is excellent in other respects and this is the only area where it falls down for me.

 

[Josh T.]

I was wondering whether it was just a problem with Epson Perfection models, as both the OP and myself found it with that scanner, and there's little if any other mention of it online.

After digging into the issue further a few days ago, it seems to be related to CCD scanners specifically, as opposed to CIS scanners, due to the particular use of a lens to transfer the image data to the CCD sensor.

Related Links:
CCD or CIS (web - by the WideTEK scanner folks)
CIS vs. CCD (pdf brochure from Paradigm Imaging Group)

So it appears that the best color capturing happens with CCD, but you also get lens distortion with that. I suppose if someone is mainly scanning documents, a CIS scanner may be the way to go (in actuality, probably just fine for photos, too). My new Canoscan 9000F MkII seems super fast (which is great), but maybe you'll want to try a somewhat less expensive CIS scanner if you decide the distortion is a determining factor. On the other hand, the cheaper scanners may not be as accurate or consistent in the Y-direction (my older CIS-type Canon LiDE 25 varies a lot, but it may just be worn out); too bad there's no way to 'try before you buy.'

I can see that it would be quite possible to correct the distortion using a corresponding opposite horizontal distortion. I'm sure a processing tool like ImageMagick could approximate it, or a sample scan could generate an exact transform.

I was curious about the possiblity of an un-distortion transformation in an editor using a scan for analysis; I'll be sure to look into ImageMagick--thanks for the suggestion.

Josh T.

 

[piasabird]

Don't try to test a 3 dimensional object in a 2 dimensional scanner.

 

[piasabird]

well different things can have different results. Keep in mind you are usually scanning through glass which can have imperfections. Also the light your are bouncing off the image may be fluorescent and coming from different angles all at once. Paper itself is not always perfectly flat either. I scan paperwork all the time. I am using a Fujitsu Fi-6140Z scanner. I noticed the lens is spring loaded. I use a sheet feeder so it works a little differently. We use a setting of about 300 dpi.

 

[piasabird]

Even something like dust could cause problems.

 

[grepnold]

Thanks Josh. I had no idea there were two sorts, and had always assumed a reducing lens and CCD setup.

This http://www.imagemagick.org/Usage/distorts/#barrelinverse is the sort of transform that might give a useful approximation, though if you wanted to map the exact distortion as scanned that might not have an off-the-peg solution of course.

I've never been aware of problems in the y direction. I can well imagine there could be (judder, uneven stepping, slack or slippage) but it's never been noticeable.