Non-linear masking techniques

by Billy Biggs <vektor@dumbterm.net>

I have been experimenting with non-linear masking techniques for digital photographs. There has been a lot of inspiration for this work. First is the wealth of terrible photographs I have taken facing into the sun or in bright rooms which I would like to salvage. Second is a lot of great recent work in colour reproduction and high dynamic range imaging which we can take and apply to traditional digital photographs.

One paper which impressed me was:

• Nathan Moroney, "Local color correction using non-linear masking", Proc. IS&T/SID 8th Color Imaging Conference, 108-111, (2000).

The paper demonstrates a simple operator based on a mask image, which should be a low-pass luminance map of the image. The operator is basically:

Where I is the input intensity, m is the value from the mask and O is our output pixel.

For example, here is an output image.

And here is the input image:

Here is the luminance mask I used:

Here is the histogram of the luminance of the original photo:

Here is the histogram of the luminance of the output photo:

And here is the histogram of the luminance map used:

How can we improve this?

The natural way to improve on this technique seemed to be to use perceptual lightness instead of luminance measures to decide how to change the image, and also to correct it. It seems wrong to change the colour of the pixels, really we just want to change the brightness.

Here are my results from experiments in using CIECAM02 perceptual attributes and transfer functions on the images. However, the results are not very impressive.