Subtract 2 identical images (B&W and colorful) to get color difference

Question

I have mostly black and white image, with some parts in color (let's say it's yellow and green). I would like to find out those colors.

For that, I suppose I might convert image to black & white, and then compare b & w image with original one to find the difference. But I'm not quite sure which algorithm I should use to do that.

Something like that in pseudocode:

image = image_from_file('image.jpg')
bw_image = image.convert_to_bw
diff_image = (bw_image - image)
# Build histogram w/o black and white parts, only color ones
diff_image.histogram

I'm primary focused on libvips to do that, but I just need an advice of how to do that in general, then I'll be able to code that.

I would also appreciate if you know any other way to do what I described above.

By black and white do you mean grayscale or binary black and white? — Jello, Feb 28 '18 at 14:56
Would this work for you? https://stackoverflow.com/a/16721373/9356981 — Adam Moller, Feb 28 '18 at 15:03
@AdamMoller thanks, that might help, I'll try to rewrite it using ruby — nattfodd, Feb 28 '18 at 15:45

score 3 · Answer 1 · answered Feb 28 '18 at 15:46

The right approach here is to convert to HSV or a similar color space that separates out saturation or chroma. Then find pixels with a high value for saturation/chroma.

The hue component of the HSV (or similar) color space then gives you the color for those pixels (yellow or green, as an angle on the chromaticity plane).

I cannot post code because you have tagged the question with two different languages, and I have no idea which of them is the code snippet you posted. I hope you can take it from here.

score 2 · Accepted Answer · answered Feb 28 '18 at 22:03

In ruby-vips you could do:

require 'vips'

a = Vips::Image.new_from_file ARGV[0]

# convert to LCh colourspace
a = a.colourspace "lch"

# Chroma (band 1) > 0 means we have some colour ... take > 10, since things 
# like jpg compression will add some colour noise we are not 
# interested in
mask = a[1] > 10

# the mask image will have 255 for TRUE pixels and 0 for FALSE
mask.write_to_file "mask.png"

The mask image might not be very helpful, but I'm not clear exactly what output you need.

LCh colourspace is useful here: band 1 (ie. C) is the distance of the pixel from the neutral axis, exactly what you want, I think.

Subtract 2 identical images (B&W and colorful) to get color difference

2 Answers2