Author Topic: Scans ain’t scans: Is there a better way of removing the “fish scales”?  (Read 1419 times)

Offline McGroger

  • RootsChat Aristocrat
  • ******
  • Posts: 2,724
  • Convicts, Commoners and Outlaws
    • View Profile
The original of the photo below (from my parents' wedding) is only small (about 2 and a half by 3 and a quarter inches) but it has plenty of detail and could be enlarged to make a good sized print. Except for one thing. The card-stock is raised, dimpled all over, so that attempts at scanning it produce what I call a “fish scale” effect over all the features. I first scanned it at home, then my son improved on it with his better-quality scanner, but light still manages to seep in, as you can see from the image.

I’ve posted a second image, a close-up showing where I’ve used the smudge tool painstakingly on Dad’s face and part of his coat. But, to strain the “fish scale” metaphor, it is a bit like trying to remove the scales without disturbing the underlying inner layer of skin, let alone the flesh. The parts I’ve worked on aren’t too bad, but they tend to make his skin look a little too smooth, a bit lumpy in parts (bruised the flesh!), or they distort some detail where I’ve not been careful enough.

My question: Is there a better way of removing the “fish scales” but without sacrificing detail?

Please have a play with the photo. Anyone wanting to colour it, Dad’s eyes were hazel, hair dark brown/black , Mum’s eyes blue, hair brown. I don’t know the colour of Dad’s clothes, but I can guess the colour of Mum’s!

Cheers, Peter. 
Convicts: COSIER (1791); LEADBEATER (1791); SINGLETON (& PARKINSON) (1792); STROUD (1793); BARNES (aka SYDNEY) (1800); DAVIS (1804); CLARK (1806); TYLER (1810); COWEN (1818); ADAMS[ON] (1821); SMITH (1827); WHYBURN (1827); HARBORNE (1828).
Commoners: DOUGAN (1844); FORD (1849); JOHNSTON (1850); BEATTIE (& LONG) (1856); BRICKLEY (1883).
Outlaws: MCGREGOR (1883) & ass. clans, Glasgow, Glenquaich, Glenalmond and Glengyle.

Offline Handypandy

  • RootsChat Aristocrat
  • ******
  • Posts: 1,581
  • Census information Crown Copyright, from www.nationalarchives.gov.uk
    • View Profile
Re: Scans ain’t scans: Is there a better way of removing the “fish scales”?
« Reply #1 on: Monday 26 June 17 10:23 BST (UK) »
I think that whatever you do to try and improve an old photo, somewhere along the line there has to be compromises. You know as well as I do that you can add stuff, remove stuff and you can adjust stuff.... what you can't do, until you get a time machine, is put extra original detail in that the photo lacks.

I've only had one shot at this pic as I know that you're perfectly able to do it yourself. I can't recall off hand which software you use but in Gimp, I've desaturated, used curves to adjust the contrast and the G'mic fine brush smoothing utility.

If I was going to try and improve it further, I would  go over some bits with about 15% smudge tool and as I went around the pic I would lightly burn in some darker bits like lips and eyes etc.

Clearly its never going to look like an original Nikon shot on Ilford FP4, but its a damn site better than having nothing  ;D

 

Offline McGroger

  • RootsChat Aristocrat
  • ******
  • Posts: 2,724
  • Convicts, Commoners and Outlaws
    • View Profile
Re: Scans ain’t scans: Is there a better way of removing the “fish scales”?
« Reply #2 on: Monday 26 June 17 10:38 BST (UK) »
Nice work, Handypandy - and so quickly! And thanks very much for the explanation. I have a copy of Gimp but mostly use PSE (an old version came with my computer; I've since updated it). I'm not familiar with the G'mic smoothing utility - must look into that. Thanks again. Cheers, Peter.
Convicts: COSIER (1791); LEADBEATER (1791); SINGLETON (& PARKINSON) (1792); STROUD (1793); BARNES (aka SYDNEY) (1800); DAVIS (1804); CLARK (1806); TYLER (1810); COWEN (1818); ADAMS[ON] (1821); SMITH (1827); WHYBURN (1827); HARBORNE (1828).
Commoners: DOUGAN (1844); FORD (1849); JOHNSTON (1850); BEATTIE (& LONG) (1856); BRICKLEY (1883).
Outlaws: MCGREGOR (1883) & ass. clans, Glasgow, Glenquaich, Glenalmond and Glengyle.

Online ReadyDale

  • RootsChat Veteran
  • *****
  • Posts: 697
  • Census information Crown Copyright, from www.nationalarchives.gov.uk
    • View Profile
Re: Scans ain’t scans: Is there a better way of removing the “fish scales”?
« Reply #3 on: Monday 26 June 17 12:13 BST (UK) »
I had a similar problem ages ago (although the pattern on the cardstock was different, I guess the logic will be the same).
Someone with more knowledge than me advised that the patterning was caused by the light from the scanner reflecting off the raised parts of the card and that a way to reduce that is as follows:
Make two scans of the picture with the second having been scanned at 90 degrees to the other. This way the scanner light is hitting different parts of the raised patterning. Then using your editing software rotate one of them so they are back in the same orientation. Overlay them (as layers in PSE) and amend the transparency % for the top one so it allows the image from the lower layer to show through (I seem to remember about 70%, but it was ages ago, so it might be worth playing around with it a bit). This has the effect of lessening what you describe as the fish scales.
I must admit, I was a bit sceptical, but given the time I gave it, it came out quite well (more time would probably have produced a better output)


Offline McGroger

  • RootsChat Aristocrat
  • ******
  • Posts: 2,724
  • Convicts, Commoners and Outlaws
    • View Profile
Re: Scans ain’t scans: Is there a better way of removing the “fish scales”?
« Reply #4 on: Monday 26 June 17 12:35 BST (UK) »
Readydale, thanks very much for your detailed description. I've got a heap of learning to do!
Cheers, Peter,
Convicts: COSIER (1791); LEADBEATER (1791); SINGLETON (& PARKINSON) (1792); STROUD (1793); BARNES (aka SYDNEY) (1800); DAVIS (1804); CLARK (1806); TYLER (1810); COWEN (1818); ADAMS[ON] (1821); SMITH (1827); WHYBURN (1827); HARBORNE (1828).
Commoners: DOUGAN (1844); FORD (1849); JOHNSTON (1850); BEATTIE (& LONG) (1856); BRICKLEY (1883).
Outlaws: MCGREGOR (1883) & ass. clans, Glasgow, Glenquaich, Glenalmond and Glengyle.

Offline Mike Morrell (NL)

  • RootsChat Veteran
  • *****
  • Posts: 502
  • Netherlands
    • View Profile
Re: Scans ain’t scans: Is there a better way of removing the “fish scales”?
« Reply #5 on: Monday 26 June 17 13:47 BST (UK) »
Hi Peter,

I read the same tip as Readydale somehere too (only the tip was to make 4 scans not 2 so the light hits the texture from all 4 directions). Here's a link to a 'how to' video for 2 scans: https://www.youtube.com/watch?v=Vy7NALaYmQ4.

On removing the (remaining) texture, the blur tool doesn't work very well for me on rougher textures. It's fine for smoothing out very fine texture but on this kind of rougher texture, it just seems to create 'lumpy' (or scaly) surfaces.

So my approach is usually to separate the details (lines) from the colours/tones on two different layers:
1. Make two layer copies.
2. Blur the first layer copy just as much as needed to smooth out the surface and remove the texture using a filter like 'Dust and Scratches', a Median filter or a Gaussian blur. See which one works best. All the lines and details will be blurred too.
3. On the second layer, apply a high pass filter to show the lines (and unwanted texture details). Add a third (masked) layer and colour it grey. You might even want a fourth (masked) layer for white. Tie these two layers to the high pass filter so they only apply to that one. Using the masks on the grey/white layers, paint over the unwanted (texture) details on the high pass filter layer. You can vary the brush and layer transparencies to let a bit of texture though so that the surfaces you paint over aren't a completely 'flat' colour. The result of the second layer (with the 'paint' layers applied) is a high pass filter that just shows the details and textures you want to keep. The 'noise' and unwanted texture has been painted out. The blending mode which seems to work best is 'overlay'. Duplicating the the high pass layer or adding more contrast to it with a tied curves adjustment can sharpen up the details even more.

So now you have blurred surfaces on the 1st layer with the paper texture removed and a 'details' layer with just the details you want to keep, It takes a bit of playing around with to get get the transparencies and blending modes just right but by using masks for both the blur and the details, everything is non-destructive and can be readily adjusted to get the final result you want.

This approach is similar to the 'frequency separation' technique used in retouching: https://fstoppers.com/post-production/ultimate-guide-frequency-separation-technique-8699

Making 2-4 scans and combining them, will avoid most of this 'correction work' altogether. But it's an approach that I've found useful when re-scanning is not an option.

I sometimes use this painting 'colour/tone (masked) layers' technique without filters just to even out the colour/tone in damaged areas of a photo. So for B&W photos the layers (matched to the photo) would be black, white, dark grey and light grey. It's sometimes quicker just to paint over large damaged black, grey or white areas than to remove individual spots, stains, tears and creases with patching, cloning and spot healing. Depends on the photo.

Hope this helps,

Mike

PS. A rough and ready example is attached. Not all that accurate. Some manual 'blur' is still needed around the eyes and mouths.
Photo restorers may re-use and improve on my posted versions. Acknowledgement appreciated.

Offline japeflakes

  • RootsChat Marquessate
  • *******
  • Posts: 7,289
    • View Profile
Re: Scans ain’t scans: Is there a better way of removing the “fish scales”?
« Reply #6 on: Monday 26 June 17 15:29 BST (UK) »
..

Offline dafydd46

  • RootsChat Extra
  • **
  • Posts: 69
  • Census information Crown Copyright, from www.nationalarchives.gov.uk
    • View Profile
Re: Scans ain’t scans: Is there a better way of removing the “fish scales”?
« Reply #7 on: Monday 26 June 17 16:30 BST (UK) »
Mike Morrell,

Some very useful information you have provided. GIMP has the Wavelet Decompose for such retouching (and it can also be used for selective sharpening).

I have used multiple scans on only a few occasions, always four at right angles to each other. Then rotate all to be the same and align them. Hugin is the best way of aligning images, but a bit complex if you are not familiar with it. In Linux I can do the alignment very quickly from the terminal - there is probably a similar method from the command line in Windows, but it is many years since I last used Windows!
Having aligned your stack of images, there are two approaches - keep the bottom layer at 100% opacity, the next layer at 50%, next at 33% and the top layer at 25%.
The other method is to use median blending of the layer stack.
Sometimes one method works best, sometimes the other.

dafydd46.

Offline Mike Morrell (NL)

  • RootsChat Veteran
  • *****
  • Posts: 502
  • Netherlands
    • View Profile
Re: Scans ain’t scans: Is there a better way of removing the “fish scales”?
« Reply #8 on: Monday 26 June 17 18:49 BST (UK) »
Good to know. Dafydd! Great that we can learn from each other in this way.

Mike

PS. Photoshop is my first 'go to' tool but I have GIMP too. Haven't done much with it yet.

Mike Morrell,

Some very useful information you have provided. GIMP has the Wavelet Decompose for such retouching (and it can also be used for selective sharpening).

I have used multiple scans on only a few occasions, always four at right angles to each other. Then rotate all to be the same and align them. Hugin is the best way of aligning images, but a bit complex if you are not familiar with it. In Linux I can do the alignment very quickly from the terminal - there is probably a similar method from the command line in Windows, but it is many years since I last used Windows!
Having aligned your stack of images, there are two approaches - keep the bottom layer at 100% opacity, the next layer at 50%, next at 33% and the top layer at 25%.
The other method is to use median blending of the layer stack.
Sometimes one method works best, sometimes the other.

dafydd46.
Photo restorers may re-use and improve on my posted versions. Acknowledgement appreciated.