weird texture on old prints

[winger]

I'm scanning some old prints - lab done, 1970s - and there's a pattern to the surface. It's small circles in rows on the whole photo. The prints themselves are only about 3" x 3" and the texture shows a lot once they're scanned. I'm guessing there isn't, but is there a way to avoid it? Or fix it?
My plan is to use Mpix to make a small book of the shots (they're from a 100th birthday party and are basically snapshots). Even in a 5x5 book, this texture will show. Has anyone dealt with this before?

 

[David Hatton]

I've seen it before on older photographs but never had to deal with scanning them. Have you tried the descreening(?) filters in photoshop?
Davidh

 

[winger]

At the moment, I still only have Elements 8, but I'll see if it has that and try it. Thanks!

 

[winger]

Well, I can't find descreening in Elements8, but I noticed that it's an option in the Epson scanning software (yes, I know, I'm still using EpsonScan instead of Silverfast - it wasn't available for mac when I first started scanning and I've gotten used to EpsonScan). Would rescanning them with descreening selected be worth it, do you think?
I was going to attach a scan, but the size requirements make it so small the artefacts don't show.

 

[David Hatton]

It's got to be worth a try. I can't think of any other way of getting a clean scan. Would scanning in oil work do you think? Somebody must have an answer!

 

[Loris Medici]

Digital reproduction with a close-up / macro lens using diffused light?

It's got to be worth a try. I can't think of any other way of getting a clean scan. Would scanning in oil work do you think? Somebody must have an answer!

 

[clay]

I'll try to put together a tutorial on this. But there is a way to get rid of all that pattern without degrading the scan. Basically, you need to download this free app, and install it. It was written by the NIH and paid for by all us US taxpayers. What it can do is some pretty sophisticated image processing. In particular, what you will want to do is do a FFT (fast fourier transform) of each color channel of your file, and then use the editor in the application to remove the big dots that represent the regular noise pattern in FK space (Frequency-wavenumber). Then do an inverse FFT on the image, and recombine all the channels. Voila - regular noise pattern is gone and your image is intact. A little involved, but much better and faster than editing it by hand. And the results are better.

One hint right now in advance of my tutorial. Change the memory usage of the Java application to something pretty big - a couple of gigs is good. It doesn't have the ability to use virtual memory very well, and you need lots of memory to do the fast fourier transforms.

I do this sort of stuff all the time in my work with seismic data. It works very well.

 

[winger]

Thanks, Clay! That sounds involved, but worth it. I'm using a relatively new iMac so I should have the memory for it.

 

[clay]

I figured out how to do this about a year ago. A close friend's wife died of cancer in 2009, and he wanted to give his daughters a nice picture of their mother when she was young. Unfortunately, one of the few pictures he had dated from 1980 and was printed in color on that silk textured paper. It was badly faded as well. I scanned the print, and immediately ran into the texture issue. I thought about using the clone tool in photoshop, but that seemed pretty daunting and destructive to the image.

As I mentioned before, my work involves using signal processing techniques, and I knew that this regular pattern was signal, and the image itself was noise, at least in a statistical sense.

I sort of grocked this whole thing out because I knew what I wanted, and just happened to run across this application that did what I needed. I googled until I found the free ImageJ application mentioned above. It allowed me to go from this to this:

Basically, the workflow amounts to this:

1) Open the ImageJ (64bit) program and go to Edit->Options->Memory and set the memory to something pretty large - I went to 6GB.


2) Open your tiff file in imageJ and look at it at 100% (image->Zoom->View 100%)

You will see something like this:

3) Then use Process->FFT->FFT.

It will think a while. Maybe more than a while. The calculations are involved. Say 5-10 minutes for a 250Mb image file.

You will get separate FFTs on each channel.

The FFTs look like this:

This the same image's Red channel. The white lines are regular patterns with periodicity - the grain pattern you want to go away.

Use the brush tool to paint a mask over these areas so that they won't appear in the image. It will look like this:

3) Now do the inverse FFT (Process->FFT->InverseFFT) on each of the channels and save each channel separately as the ImageRED, ImageGREEN and ImageBLUE tiff file.

4) Load each of these channels into photoshop at the same time and merge them. This is done using the Merge Channels selection in the Channels pallet. You will have to look in the photoshop help menu about how to do this, but it is pretty straightforward. Merge the files into the appropriate channels and you have your image back without the noise.

 

[glhs116]

That, my friend, is impressive and incredible!

 

[Loris Medici]

Wow, thanks much Clay - I'll keep that in my mind...

Regards,
Loris.

 

[David Hatton]

Wow!
Many thanks for that.Amazing process.
Davidh

 

[winger]

That, my friend, is impressive and incredible!

Very! As soon as I have the chance to get some time on my other computer (without the baby in my lap), I'll try it out!