Help:Scanning

From Wikimedia Commons, the free media repository
Jump to navigation Jump to search
flatbed scanner
flatbed scanner

Scanning an image or document for Commons can be relatively easy, if you know what you are doing, and, if you have an interest in history, it can be an excellent way to share your interest with the world.

Excellent sources for public domain works include:

  • Libraries (in particular large municipal libraries and, even better, university libraries, where students and other authorised users generally get freer access to the older books than is normally granted to the general public)
  • Historical societies
  • Second-hand bookshops
  • Yard sales and charity shops

General advice

[edit]

Check your monitor for proper adjustment, especially the brightness and contrast settings. Too bright or too low-contrast, and your images will tend to have a grey tinge to them on better-adjusted monitors. Ideally, you should be able to see three circles on this image, which tests the monitor calibration. Commons:Image guidelines#Your Monitor provides further advice.

Unless your scanner is not capable of it, never scan at less than 300 dpi[1] File sizes may be large, but for engravings, paintings, and illustrations of any complexity, this is about the minimum resolution required to reproduce them at a reasonable quality.

400 dpi is a good general resolution. For engravings and similar works, though, 600 or even 800 dpi is useful, as master engravers (such as William Hogarth and Gustave Doré) have often included details smaller than the naked eye can make out. For images up to about the size of a postcard (about 3" by 4" / 8 cm by 10 cm), 500 to 800 dpi may be convenient, as it allows them to be scaled up beyond the original size to a certain extent. Similarly, 600 or 800 dpi is a good choice when scanning from rare works – the extra quality will be appreciated. Be assured that 1200 dpi is almost always excessive, except for slides or microfilm. Consult your scanner's manual in such cases.

Clean the glass before scanning, especially if you have pets. — Hair, dust, and the like tend to get on the scanner bed.

Use the preview option of your scanner software (if available) to place the image as straight as possible before you scan. Images can also be rotated afterwards, but this is more time consuming. Also, every scanner software is different. So play around with all the functions of your software (switch to “Professional” or “Advanced” mode if given the choice), until you know how to use them well.

Make sure that the object to be scanned lies fully flat. If it is only one sheet, put heavy books on it. If you are scanning part of a book, push down firmly on the cover with your hand while scanning. Obviously, this does not apply to fragile works.

If at all, autolevels options should be used with some caution. Compare the preview with the original (as best you can, given the original is being scanned), and check if the results make sense. If you have any experience with image editing, or think other experienced people might help you later, then turn autolevels off.

Don't worry about large works that don’t fit on the scanner in one piece: At graphics labs, such as en:WP:GL/IMPROVE, Commons:Graphics village pump and Commons:Graphic Lab, there are usually people who are quite happy to stitch together an image from multiple scans.

Tip: It makes things much easier for them if you use the edge of your scanner's glass to straighten whatever you are scanning against, so that all partial image scans are taken at the same angle. However, if you can't do this, they can usually cope. Also, scan the parts at a fairly high resolution (600–800 dpi are often good): it will make it easier to conceal the joins between the images if it can be scaled down a bit afterwards.

Be sure to turn off “autolevels” of your scanner, no matter how good the results are normally: two consecutive scans with autolevels off will be a good match in paper tone and colours, and thus much easier to "stitch" together. When turned on, you can end up with subtle or even major differences in tone that make the stitching together more difficult.

Files on Commons can be no larger than 100 MiB. This is almost always ample for both PNGs and JPEGs of up to about 800 dpi resolution at any size that will fit on even the largest scanner. TIFFs can be much larger, but, as explained below, it is preferable not to upload TIFF files anyway.

Reducing bleedthrough

[edit]
printing from the backside visible through the paper
bleedthrough is reduced by scanning with black paper behind the scanned page

Already with standard paper thickness, but especially with thinner paper, you can often see the printing from the backside or the next page through the paper. This can be greatly reduced by scanning with a blank paper (or two, if necessary) behind the scanned page. If the printing from the backside shows through, you should use black paper for this, if it is only the next page, white is preferable.

The black paper behind the page causes a slightly darker white point that can be compensated for later.

If a scan does show shining-through text and you can't rescan it as described above, it can sometimes be corrected with little loss using image manipulation software. A technique that works quite well for grayscale images is explained at Commons:Pearson Scott Foresman.

Color calibration

[edit]
IT8 target

Manually changing scan settings can help improve the color accuracy of your scans. For better results, you can buy an IT8 target, which is a sheet of defined colors that is scanned and then analyzed by software (such as the free LPROF) to create a custom color profile for your scanner.

PNG vs. JPEG

[edit]

Please note. The following advice refers to scanned images. If an image has originated from a digital camera, it will usually be in JPEG format. There is no point in converting an image which is already in JPEG format to PNG unless you intend to do extensive edits to the image and want to be able to save your work as you go along. (Saving as JPEG repeatedly can increase the amount of JPEG artefacts significantly.)

PNG saves images using lossless methods; GIFs and JPEGs (sometimes called .jpg because of a DOS file name limitation) can add artefacts (ugly errors, pretty much) to your picture. In general, GIFs are mainly used for animated images, and JPEGs and PNGs are the main choices for still images. As most scanners are not set up to capture moving images, let's concentrate on PNG and JPEG.

PNG is a safe bet in most cases, but if a PNG is very large (more than 12.5 million pixels, or more than roughly 4000×3000), Wikimedia software can't display the image. A full-colour PNG file can also be quite large, although the increase in maximum upload size to 100 megabytes helps a bit here. Programs like Optipng or PNGcrush can help make your PNG files smaller without any loss of quality.

In any case, it's usually best to scan to a lossless format like PNG, TIFF, or, if necessary, BMP first. A JPEG file has already lost quality, and, with some settings, may have lost a significant amount; switching to PNG will not bring that back. In addition, artefacts accumulate each time you edit and resave JPEG files. By starting with a lossless format, even if you have to go to JPEG in the end due to size issues, avoids unnecessary loss of quality.

Best practice: When possible, upload a PNG file as a lossless archival version, even if it can't be displayed on Wikipedia because of its large size: You can always upload an additional JPEG version, and add links between them in the "other versions" section of the upload template. This way, any future image manipulation can be based on the lossless PNG file, instead of a lossy JPEG.

In better image editing programs, you can trade off quality vs. compression when saving to JPEG format. In general, use the maximum quality 100 – the scale is 1 to 100 with 100 being best quality. In Photoshop, which uses a scale of 0 to 12 instead, use 12.

If the compressed file size exceeds the current limit of 100 megabytes, consider either reducing the quality or, if the material is of historical significance, requesting assistance from the Commons:Graphic Lab. If the quality is reduced, check the image at full resolution before uploading to make sure it still looks okay. This old version of Sadko.jpg appears to be made up of thousands of little squares when viewed at full resolution, which is one of the things that can happen if the quality is set too low. The current version has twice the file size, but avoids the worst of these problems.

If you have the choice, choose to save PNGs with the "smallest file size" or "highest compression": The PNG compression algorithms are entirely lossless, it just takes longer save the more compressed ones. In practice, the much smaller filesize makes it so much more efficient that the time spent dealing with the compression doesn't matter.

PNG vs. TIFF

[edit]

PNG has much smaller file sizes than most archival TIFF. Scan to PNG if the scanner software has this option, and otherwise consider converting to PNG afterwards.

TIFF is offered as a courtesy to any museums or other archives that want to upload their files. However, TIFF can contain almost any other format — including, in theory, lossily compressed formats like JPEG! Therefore, if you need to use TIFF, make sure you choose a lossless compression, preferably LZW or Zip/Deflate.[2] Baseline TIFF encoders will always produce lossless TIFFs.

Editing your image

[edit]

Always upload your original scan, preferably as a PNG or PNG/JPEG pair. This way people can clearly see what your manipulations were, and allows other editors to fix things if the edits accidentally damage the image.

Common manipulations include:

Levels adjustment
Adjusting colours to match the original.
Cloning out hairs
If you have cats, there is a high probability that at least one hair is present on your scanner.
Attempts at restoration
Fixing tears and stains.

These are beyond the scope of this help page. Contact an experienced user for help with these.

Black and white, grayscale, or colour scanning?

[edit]

If your image is in colour, it should of course be scanned in colour. With black-and-white images the decision is a little more complex.

Binary black and white is usually not a good idea — a grayscale or colour scan tends to look a lot better because it shows oblique edges without the jaggedness of binary black-and-white, since it allows anti-aliasing to smooth out pixelation. However, there's something to be said for both the other choices.

float
float

When scanning from a reproduction – I could only bring a photocopy of this newspaper to a scanner – it makes little sense to preserve the paper texture. This image is scanned in grayscale, with the contrast adjusted upwards to give a smooth, white background, and to get the main parts of the lines to pure black. This puts the emphasis squarely on the picture, and since the lines that make up this image are fairly thick (for an engraving: the smallest are about as wide as a line from a ballpoint pen, and are all visible to the naked eye if you look closely), such adjustments would not result in any real loss of detail. This way, the image can be printed without trying to reproduce a paper texture which the paper it is printed on will have anyway.

float
float

This scan is from an original, and is a different type of engraving – a copperplate engraving instead of a woodcut from a newspaper. In this one, some of the lines are so fine that they are barely visible to the naked eye (at its original size), the ink is slightly tinted from age, and the paper has a nice feel of oldness to it. Some of the detail in the very delicate fine lines might be lost with a lot of post-processing, and the ink and paper add to the interest of the piece, so this one is best kept in colour. However, it is somewhat harder to print this one.

If in doubt, try both ways, and then decide which one you like best. Note, however, that you can go from colour to greyscale, but not the other way around. So if scanning something quite rare, colour is probably best.


Half-toning

[edit]
Left: Halftone dots. Right: Optical impression at sufficient distance.

Half-toned images are used in most modern printing. They consist of an array of dots at regular intervals, with the size of the dot determining how dark it is. Unfortunately, half-toning can look awful if you zoom in too far and may create disturbing Moiré artefacts. Consider this image:

float
float

The original was made by using engraving for the black lines, followed, as I understand it, either by hand-tinting or an additional plate for each colour. However, this version was clearly scanned from a modern book, and at full-view, all its half-toning dots are visible.

If possible, try to use the original sources. Of course, this is not always possible. So if your work is half-toned, but still under a free licence, please do scan it for Commons! At least if scanning resolution was high enough, most half-toning can be fixed with a little manipulation afterwards, and, even if the image ends up, by necessity, at a lowish resolution, it still shows things that would otherwise not be available for Wikimedia projects.

“Remove moiré”, or “descreen” functions of scanner software make a start towards fixing half-toning. But usually the results are not on par with some of the available descreening software. Those functions may be more destructive to the image and may also prevent better descreening software from removing the remaining half-toning artefacts afterwards.

A half-toned image cannot have more detail than the distance of the dots that make it up. So if your work is half-toned, it's best to manipulate it in a image editing program afterwards. To do this, it is desirable to use an oversampled scan (a higher resolution than that required to reproduce the smallest details of the actual image). You may want to use a resolution that makes it possible to distinguish the individual printing dots.

Merging dots with Gaussian blur is straight-forward to do and utilizes functions available in most image-editing software, but it destroys more detail than necessary. Since the printing dots have to actually merge half-way into each other until the raster disappears completely, you end up with only up to half of the original resolution.

You need specialised software to remove dot patterns in the frequency domain, but this way you can retain almost all the detail of the full original resolution.

Software is available from Cornell University and Picture elements to automatically fix black and white halftone images, if scanned at 600 dpi.

Note: The techniques used for halftoning have changed over time. On early half-toned images (circa 1890–1920) these methods will probably not work.

Merging dots with Gaussian blur

[edit]

Note: This can work very well in some cases, if done carefully and if you do not need more than half of the original resolution. However, it cannot be undone, so please upload an unblurred copy as well. Also, note that this should not be done to lithographs or other such images.

Gaussian blur can be used to smooth halftone images while destroying 50 % of sharpness. By using FFT instead, almost all details can be preserved.
Up: Halftone picture.
Middle: after applying a Gaussian blur with σ = 2.
Below: Dot pattern removed in the frequency domain.
(results seen better at full size)

First (perhaps after carefully correcting the white-balance) use the Gaussian blur filter with a radius just big enough to make the dot raster disappear. Now you might want to maximise the contrast and maybe do another tonal correction and other stuff. Then you can scale down the image by factor of the blur radius. Because there are no smaller details left, this won't hurt much.

You might want to use a resampling technique that retains more sharpness instead of having to sharpen afterwards.

In binary black-and-white half-toned images, however, you may be able to get away with just blurring it a bit, then using the sharpening filter and increasing the contrast. You should probably scale it down a little afterwards, but with practice this can salvage a black and white half-toned image to good effect.

Removing dot patterns in the frequency domain

[edit]

You need the G'MIC plug-in for GIMP, which includes both a Fourier transform filter and an additional descreen filter that automates the whole process and ensures that you don't miss anything.

(See here for step-by-step instructions on how to manually perform the steps from the automatic descreen filter.)

This works incredibly well if the patterns to remove really are coherent throughout the image. If there are (small) areas where the patterns are absent, the pattern that is removed from the other parts will show up here instead. In this case you may want to load the previous image as a layer underneath the descreened one and merge in the undistorted areas from there, e.g. add an alpha channel to the upper layer and selectively let the lower layer show through using the eraser tool.

Similarly, if the scanning resolution is so high that the printing dots are completely separated, then the pattern removal may work less. You may have to try to feed several resolutions of the image into the frequency transform. Initially scanning the image at even higher resolution may still be helpful to reduce image noise.

You may then want to reduce the resolution to get closer to the original resolution. It may be a little difficult (compared to above blur technique) to find the right reduction factor. You can try to find the minimum blur radius that completely removes the pattern from the original scan, as described above under "Merging dots with Gaussian blur" or use the measuring tool to determine the pixel distance between two adjacent printing dots.

Now reduce by at maximum half of that factor. (Due to the sampling theorem, bitmap graphics require twice as many pixels per dimension than the maximum frequency [number of details] to reproduce.) Before resampling you may want to apply other filters such as a median filter (“despeckle” in GIMP, or the less destructive pixel denoise from G'MIC) to (some parts of) your image.

[edit]

Engravings are, perhaps, the easiest type of art to work with, and, if you have access to a good library, 19th century illustrated newspapers were common, often had very good quality engravings, used quite a lot of them, and are often fairly-well preserved.

There are two main types of engraving:

The first is to make it out of individual lines, as in this (originally approximately 2" / 5 cm tall) small engraving of Charles Dickens from the Entr'acte, a Victorian theatrical newspaper. This technique is also used for far more complex drawings, for instance:

If you look at this image of William Hogarth's Gin Lane at full size, you will see that all the shading, all the detail comes down to fine lines and crosshatching. The fine lines are actually invisible to the naked eye, instead blending into shading.

This is perhaps the most common form of black and white engraving.

Now consider this engraving:

Technically, this is actually not an engraving, but an etching. An acid-resistant coating was put over the plate, then areas were scratched away to allow acid to get at and texture the plate. The longer the acid is in contact, the rougher the plate's surface gets, and so the more ink it holds. By using several baths, changing what is covered as you go, you can create delicately-shaded works such as this one, with the shading made up of a sea of irregularly-shaped pits.

Etching generally cannot get as much detail as an engraving proper, as a certain amount of randomness comes into play from the acid pitting the surface irregularly. An etching is inherently "noisy", with irregular dimpling of black and white, as it's altering how much "noise" there is in any one area that actually makes up the art.

This distinction matters to scanning: In a scan of an engraving proper, every line should be distinct at full resolution, unless the engraving is extremely large, but in an etching, the artist did not physically choose the exact texture that creates the colours or grayscale, so a slightly lower resolution is fine. If you have a choice, somewhere between 300dpi to 800dpi is a good choice, and always go on the higher side for copperplate engravings – the details in a copperplate engraving can literally be microscopic.

A good scan of engraving, etching or similar should:

  • Generally speaking, have a minimum resolution of 300 dpi.
  • Show every line that makes it up distinctly, if an engraving. In an etching, it's basically made out of noise/static/irregularly shaped pits, with the location not precisely chosen by the artist. Just scan them at a reasonably high resolution, and make sure all graphical elements are visible.
  • If it's a black and white engraving, and you've decided not to show the paper texture, adjust the levels so that the background is smooth, pure white, and the ink (at least where there's plenty of it) is a nice dark black. If you are scanning in colour, still make sure the paper is reasonably light in colour, and black areas do not look washed out, but reasonably black. This will make it look far better when scaled down for viewing on Wikipedia and other projects.
  • For colour engravings, see also the advice of the next section.

A note on woodblock engravings

[edit]

Woodblock engravings, particularly from Victorian periodicals, often contain fine white lines that show the divisions between the woodblocks that were glued together to make the full image. (Example Image:Design for an Aesthetic theatrical poster.png is fairly cleanly divided into four smaller rectangles.)

There are multiple views on whether it is best to edit the image to remove them or to keep them in for authenticity. Graphics labs, such as en:WP:GL/IMPROVE, Commons:Graphics village pump and Commons:Graphic Lab are probably the most useful places to go for restoration work; describing how to do extensive restoration work yourself is probably out of the scope of this tutorial.

Paintings, full-colour illustrations, and similar works

[edit]

The methods for scanning full-colour illustrations, paintings (however, see below in this case), and similar are not greatly different from engravings, but it's best to adjust the colours afterwards to make it look as much like the original as possible.

  • Scan at a minimum of 300dpi.
  • Using a graphics editing program, adjust the levels, brightness and contrast, and so on, until the colours are as similar to those in the actual picture as possible. Keep a copy of the untweaked scan, and compare it with the final version to make sure you haven't accidentally messed something up. Also, this was said in the general advice section already, but make sure your monitor is appropriately calibrated, as described in Commons:Image guidelines#Your Monitor – otherwise, what looks realistic to you and what looks realistic to everyone else will be different.

A warning about paintings: For paintings done on a canvas (e.g. most oil paintings, acrylics, and so on, in most cases, it's not going to be possible to get the original to a scanner, and, if the painting is old, it might damage it even if you could get it to one. If, however, it is possible, and damage is unlikely—e.g. a painting you've just made yourself, hence in good condition, note the texture of it. A little texture is fine, but if some parts stick out much more than a couple millimetres from the surface, you are probably best photographing it.

In many cases, though, you'll be scanning a painting from a modern reproduction. This can lead to mixed results. In lower-quality reproductions, you'll be dealing with half-toning, as described in the earlier section on it. Use the advice given there to attempt to ameliorate it. However, really good reproductions, as can be found in some high-quality art books may not have half-toning, or have it so fine that it doesn't matter except at the most ridiculously high of resolutions. In these cases, scan it at at least 300dpi then adjust it in a graphics program as described for scanning from an original painting.

As always, Graphics labs such as en:WP:GL/IMPROVE, Commons:Graphics village pump and Commons:Graphic Lab can assist you if you find this difficult. Also, check the copyright status first. Bridgeman Art Library v. Corel Corp. and similar rulings in other countries mean that, in most cases, if the original is in public domain, a copy is as well.

However, note that the United Kingdom has unusually strict copyright laws that may protect a heavily-restored image produced there. If in doubt, Commons:Licensing attempts to explain the full rules related to copyright, and Commons:Village pump may be able to help you if you are still uncertain.

Cropping

[edit]

Try and leave a little whitespace around the image when you are scanning it in full. This makes sure you don't accidentally remove useful parts of the image or of its caption, or give the impression you have. Obviously, this may not be possible if the image goes right to the edge of the paper, but putting a piece of blank white paper behind it can help. (Don't worry about this when using black paper behind to prevent bleed-through.) Scan the image in multiple parts, if necessary—as mentioned in #General advice, support is available to stitch an image together from its parts.

When giving a detail from a larger image, try and trim it so that distracting details you do not intend to draw attention to are minimised in visual effect. For example:

This is a detail from a Punch cartoon—this Punch cartoon, in fact—that was being cropped for the English Wikipedia article on Gilbert and Sullivan. As such, The main image of Sullivan, and the tiny W. S. Gilbert were the important parts. Part of someone who is probably F. C. Burnand can be seen in the upper left-hand corner, but the crop avoids showing his face, so it doesn't attract too much attention. This detail is also from the lower-right corner of the original, so it's fairly sharply cropped on the right and bottom to avoid including (most of) the black line that frames the image as a whole, as having a thick black line on only two sides of an image would unbalance it. Serendipitously, the tiny bit of the black line that got left in on the lower edge and the bit of Burnand's moustache in the upper left completes the frame, creating a nice, even rectangle.

Caption

[edit]

If a picture has a caption written in the same medium, it is better not to crop it, so that the information it may contain (original title, publisher, date, etc.) is immediately verifiable. Note that it is possible to produce a cropped version from the full scan with caption, while the reverse is not. Always provide it with the captions, then upload any crops as a separate image

See also

[edit]

Notes

[edit]
  1. Scanners use dpi (“Dots per inch”) to measure the resolution of the proposed scan. At 300 dpi, a 1 inch by 1 inch section (2.54 cm by 2.54 cm) of the original image becomes a 300 pixel by 300 pixel section of the scan.
  2. More on the topic: https://photo.stackexchange.com/a/69661/45210
[edit]