Re: [Gimp-user] Captioned newspaper photo

[Liam R E Quin <liam holoweb net>]

On Mon, 2018-06-11 at 22:05 +0200, lkl316 wrote:
> I have a captioned newspaper photo iI would like to clean up and make
> a photo on
> photo paper. It has some creases I would also like to repair. Has
> anyone
> attempted this and give me some guidance?

i've done a lot of this, and there's even books on the suject, although
only  for PhotoShop™ and not GIMP as far as i know.

For example, i have Digital Restoration from start to finish, by Ctein.

Some rough guidelines without seeing what you have.

First, scan the image. Put solid black or green paper behind the single
sheet of newspaper to reduce show-through from the other side of the
paper. Use at least 1200dpi for the scan if you can. If it's black and
white you can scan in greyscale but try and get a good range of levels,
and use 10 or 16 bits per pixel.

Then, in GIMP, if it's a black-and-white picture, you can use
colours/levels/auto (at the expense of losing a tiny bit of detail) or
colours/curves, until you have black in the very darkest place and
white in the brightest.

Now, since you said it's a photo (as opposed to an engraving or
drawing), it'll be screened - lots of tiny dots.

You can decide to keep the dots, but then when you repair the tears and
marks with cloning you'll need ot be careful to keep the pattern of
dots perfect. It's doable but takes care and lots of control-Z (undo).

Or you may decide to get rid of the dots.

The simplest approach to get rid of the screen dots is to use
filters/blur/gaussian blur. You can use the Measure tool ("M") to
measure the average number of pixels between centres of dots - e.g.
count 10 dot-gaps and divide by 10 - and that's likely to be the radius
you'll want for the blur. If in doubt, copy a small pice of the picture
into a new image and experiment as it'll be faster.

After doing a blur, use image->scale and either cubic or linear
interpolation, and scale down to 10% of the original size. If your blur
radius was large enough you will not see any dots or patterns. You can
now use levls and/or curves again to increase the contrast and range a
bit, and then filters/enhance/unsharp mask... don't over do the unsharp
mask. if unsharp brings out the dots, use Undo, go back, increase tbe
blur radius, you want the minimum blur that gets rid of the pattern.

Note: there are more advanced ways to do the dewscreening using g'mic
and/or frequency decomposition but the results aren't usually much
better for newspaper images.

Once you've established that you can downscale the image without
introducing weird patterns, undo that and scale down to 20% instead, so
the image is twice as big as you want it.

Now use the clone tool with a soft round brush to work on the creases.
You may also need to select darker or lighter regions with the free
select tool, feather the selection by e.g. 30 pixels, and use curves,
to get rid of rust spots, fungal stains, water marks, lizard poo and so
forth.

If there's lizard poo you should disinfect GIMP after use :)

If you don't have a soft round brush make a new one, then in the tool
options for the clone tool set hardness to 85% or so, and radius e.g.
to 15 pixels, you can experiment. Depending on your keyboard, { [ } and
 ] may change the brush size so you don't need to look away from the
canvas.

For large areas it can be less work to make a "patch" from another part
of the image with select/feather/copy/paste.

Finally, scale down by 50% and use levels and/or curves and then
unsharp mask. Don't over do the unsharp mask.

Aim for 144dpi for the printed version, if possible, or 72dpi at the
lowest, in greyscale -- newspapers are printed with a 75-line-per-inch
dot screen usually so you're not going to get better than that.

The reason to scan at such high resolution is that the exact size of
each dot varies according to the image value in that region, so that
details in the original photo are hidden in the sizes of the dots. The
reason to use 16 bits per pixel when you scan is firstly that the
intermediate values around the edges of the dots are part of that
detail, and secondly that otherwise you can end up with very few
distinct colour values in your scanned image, and this can lead to
visible artifacts when you process it. If you went the blur route
that's less of a problem, but if you decided to keep the dots it can
make the cloning job really hard.

If the image is in colour it can help to separate the scan into CMYK or
RGB layers and blur them independently. You can also sometimes improve
print registration/alignment errors this way.

Liam/slave ankh


-- 
Liam Quin - web slave for https://www.fromoldbooks.org/
with fabulous vintage art and fascinating texts to read.
Click here to have the slave beaten.