Monday, February 15, 2010

Fixing Silvering in Old Family Photographs

On Friday my wife handed me some old photos to scan.  One taken of her great-grandmother exhibited what has been called "silvering" - the dark black areas have become very shiny and silver-colored.  Scanning such an image on a flatbed scanner works poorly.  The angle of illumination causes the silvered areas to come out very light when they are supposed to be the deepest blacks.  Sometimes you can re-image the photo with a digital camera instead, if you can get the lighting at just the right angle to avoid specular reflection and use a polarizer.  But then you have to deal with focus, barrel distortion, uneven lighting and all kinds of other issues that scanners solve for you.  And you still have dark areas that are just too light.

So I just scanned it to see what would happen.  My working size was about 32 megapixels, but here is a reduced version of what I got.  The silvering has really ruined a nice portrait.

Hmm... what can I do with it using cheap Corel Paint Shop Pro?  Well, I noticed that the silvered areas were basically blue, which made them distinct from the rest of the image, so I increased the blue saturation then split the photo into its blue, green, and red channels, generating new black and white images for each channel.  The red channel image was actually not too bad by itself. On a photo with moderate silvering, just taking the red channel may be enough.

I colorized the red channel to match the sepia tone of the original, then arithmetically subtracted the red channel from the saturated blue image leaving an image with just the silvered areas as blue.  I then subtracted the image with the blue areas from the original.  I fiddled with the blue image quite a lot to get the blue hue and saturation to match so that the subtraction worked well.  I finally converted the resulting image to gray-scale to eliminate some color mismatching in the background silvering (it was actually more green than the silvering in the hair and collar) and then re-colorized it as sepia.  A little more fiddling with contrast, brightness, and a little smoothing gave me the final image.

I think it turned out rather nice. The silvering is gone, and a surprising amount of detail emerged in the hair.  You can click on any of these images to get a little bit larger size.

7 comments:

  1. Can you please explain what you mean about arithmetically subtracting the red channel from the blue? I was with you up until that point. Thanks.

    Beth

    ReplyDelete
  2. Hi Beth,
    In most image/photo processing software, there is a way to split the image into red, green, and blue channels. And there is usually a way to subtract, or add, or difference these channels on a pixel-by-pixel basis. For example, in Paint Shop Pro IX, the command is Image, Arithmetic. It brings up a dialog box where you can select which two images you want to do an arithmetic operation on. In this case each pixel value in one image is individually subtracted from the value of the corresponding pixel in the other image. So subtracting the blue "mask" I created from the original, pulls the blue silvering out of the original.
    Scott

    ReplyDelete
  3. For a more generally applicable solution for the silvering problem check out the color deconvolution plug-in on 4N6site.com. Silvering examples are shown on:
    http://www.4n6site-com.nl01.members.pcextreme.nl/improc/decoplugin/examples.htm

    ReplyDelete
  4. Amazing. Thank you. I am going to try this.

    ReplyDelete
  5. I just tried to follow your instuctions and also got lost. You should make a Youtube explainer video and tell us how you did this. Can't find ANY GOOD INFO on the web about removing "silvering" or "silver mirroring". Your results are the best/only ones that I have seen that are excellent. Please do a video!!

    ReplyDelete
  6. I don't get your intermediate step, the "Increased the blue saturation" How did you do that?

    ReplyDelete
    Replies
    1. Depending on your software, there should be a adjustment called red, green, blue, with controls to increase or decrease each one.

      Delete