Contrast Masking
Imaging - Imaging
Monday, 24 December 2007 01:16
We all know the situation - a picture looks unusable due to extreme contrasts with either blown out highlights or many deep dark shadows ... or both. There's a technique that may help to rescue such pictures: contrast masking. Contrast masking isn't exactly new - it has been a well-known technique with film for decades - it got much easier with with imaging application such as Adobe Photoshop or The Gimp (freeware).  Besides some filters the imaging application should support layers as a pre-requisite.

Ok, enough of the babbling.Here's a quite extreme sample picture which has obviously plenty of potential for corrections ....

The workflow for contrast masking is as follows:
  1. make a copy of the original image
  2. remove all colors
  3. inverse the image (= B&W negative)
  4. Gaussian blurr the step 3 image - this is your contrast mask
  5. overlay the mask onto the original (as a layer - layer 1 = original, layer 2 = mask)
  6. change the opacity of the mask layer down to a "sufficient" degree - 20% may be a good start here
  7. adjust the levels of the resulting image to recover deep black and bright white
  8. done
You may wonder about step 4 here. Why should you blurr the "shadow" image ? Well, try it. Without blurring you'll notice that the image will be rather soft. By blurring the negative the details of the original image are retained.

Sounds complicated ? It isn't -it really isn't.

Here's the transition of the contrast mask during the workflow ...

B&W (step 2)
B&W Negative (step 3)
final mask (step 4)

Let's apply the contrast mask to the original image now ...

original image overlayed by 20%
with the contrast mask (step 5 + 6)
final result after levelling (step 7)

Let's compare the original image with the processed one ...

You will notice the improvement immediately - the shadows are much less pronounced and the sky is a bit darker now. Nonetheless the processed image still contains deep black as well as bright white. The difference is also obvious in the histograms - the processed image shows a smoothed fading towards the extreme ends of the histogram and the primary data has been shifted towards more healthy mid-tones (to some degree).