cleanup staining on printed matter

[bugbear]

I have an old tool manual, with stains from aging
and oil (I guess it was in a toolbox).

To all intents and purposes I'm trying
to recreate the original (shiny and new)
printed document using the existing
old document as a starting point.

Since the content is black, and high frequency,
and the "problems" are yellowish and
low frequency I'm thinking fouriers, blurs
and colour space manipulations
ought to work, but I can't get the
combination right..

Any ideas and assistance gratefully
received.

The main tools I have access to (on Linux)
are gimp, netpbm and ImageMagick

BugBear

[CJ Swartz]

Welcome, Bugbear, we have one or more Gimp users (or former users) on board who may be able to help you better than I, but I can show you what I did using Photoshop. First I tried a number of layer adjustments to your RGB image, but didn't know if Gimp has "Selective Color", and "Channel Mixer" so decided that going grayscale would be easier. After changing the image to grayscale, I used a Layers adjustment to clean up the whites and darken the black ink. To erase the black marks around the border, just make a rectangular selection inside the page with most of the smudging outside the selection, change to inverse and Edit->Clear the selected smudged areas back to stark white. Then you can use an eraser tool to clear up any small marks left on the page.

Hope this helps, or that a Gimp user will be able to give the steps for using that program.

[jeaniesa]

Similar to CJ, used the red channel to change the image to grayscale. (The red channel showed the least amount of damage, so it makes for the easiest starting point.) I then applied a Levels adjustment (I'm pretty sure Gimp has something similar). I pulled the highlight slider way to the left to get rid of the darkening caused by the stains. Then I pulled the shadow slider to the right to darken the text and diagrams.

I ignored the edges - that's a simple matter of cloning or cropping away. I also didn't sharpen, but you might want to consider doing that after you've removed the damage as outlined above.

Jeanie

[Stephen M]

Bugbear - this is a 'lineart' image, so firstly I would scan at the highest optical resolution that the scanner is capable of. Place weights on the book if you can't cut it up or keep it really flat. Also look into backing papers to reduce bleed through of content from the backing page. Where possible I will attempt to scan straight rather than rotating in scanner or image editor.

So firsty try a direct bitmap/lineart scan if you can instead of gray or colour. Also try those and compare the quality of the various cleaning approaches vs a direct lineart scan. Please evaluate on test prints or whatever similar final media to the end job. With lineart type data, what looks crappy onscreen may print nicer - or it may output just how it looks!

I will leave it up to you to read between the words and explore the options in your software. Below is some generic image processing info which should be suitable to many apps.

Another cleaning option is to use thresholding to clean up the image. This can be done with a conversion from grayscale 8 bpc to bitmap 1 bpc data (and then back to gray or RGB) or by staying in contone modes of gray or RGB and using a threshold command starting at 50% (127) or adjusting to suit.

For a 50% threshold, any tones under 50% will become solid white, any tones over this value will become solid black. Next some anti-aliasing is added to the image (very very small blur or many, many other methods to mung the data to produce averaged edge pixels) and or using levels/curves and sharpening to clean up the edges. Other options include resampling the resolution down a bit from the higher scan size resolution that may be higher than needed for output (try single step, 10% increments or 66% or 33% steps when resampling to see which effect is best for the content at hand).

A less global approach is to use an edgefinding filter to make a selection or mask so that thresholding only affects low frequency data (not the sharp lines, just the flat areas).

Regards,

Stephen Marsh.