Matching Colors of Source Image to Target Image

Hi. I want to match colors of source image to target image.

Source and reference images are identical with same resolution, aside from the color. (Actually, target image is a graded version of source image)

Here is the source image:

Here is the target image:

So far what I did is:

• -Converted source and target images to LAB color space

Now, I'm not sure how to proceed from here.

• -Shall I go with histogram matching in LAB (not sure how to do it)
• -Calculate deltaE between each pixel, calculate mean etc (also not sure how to do it)
• -Use any other Color Space such as LCH

Please help me regarding this...

[Walter Roberson]

What do you mean by matching colors in this case? Are you looking for a linear transform in rgb, or in LAB? Looking for a curve of hue correction? A look up table of original and target colours?

Is it assumed that any given source color is always matched to the same target color?

[cona mx]

Yes exactly, my aim is to create a look up table of original and target colours.

Answer 1

[Walter Roberson]

[old_colors, old_idx] = unique(reshape(Original_Image, [], 3), 'rows');
reshaped_new = reshape(New_Image, [], 3);
new_colors = reshaped_new(old_idx, :);

The lookup table is then that each row of old_colors is matched to the corresponding row of new_colors .

This assumes that any given source color is always matched to the same target color.

[cona mx]

Thank you. Let me check this.

[cona mx]

Does this code work for images converted to lab?

[Walter Roberson]

Yes.

Note: rgb2lab() always produces double even for uint8 source, and lab2rgb() of that will produce double, with the round trip being equivalent to im2double(). Also I notice that the result of lab2rgb can be slightly negative due to round-off error such as for (0,242,65). However, if you im2uint8() the result of the round trip, then that will be identical to the original RGB.

[cona mx]

okay, thank you very much for your patience and support.