GoktugEk
commented
2 years ago Thinking to use Delta E metric, introduced in 1976, and significantly improved until 2000s. It is a metric that calculates the distance to a metric between 0-100(0 is identical, 100 is completely reverse).
Calculating the distance between two pixels can be done with Delta E. I am thinking to use Python and Colour(https://colour.readthedocs.io/). Calculating the average, or the norm of the difference might makes sense. Average makes more sense to me right now.
After a canvas has been processed, we would like to compare its similarity to an image. We will aggregate the results of pixel-wise comparisons for the overall image similarity.
The algorithm should be answering the question correctly;
Given an human oracle and 3 images, the algorithm should mostly be able to correctly compare pairwise similarity of those images. Formally, the similarity relation ordering(sim(i_1, i_2), sim(i_1, i_3), sim(i_2, i_3)) should always align with the oracle's answer.
We will only consider pixel-wise comparison algorithms for this. A nice introduction should be color difference page.
Also check this cow thread
@GoktugEk is responsible for this. Goktug please provide the results of a brief research followed by a prototype implementation in your language of choice. Assume initial pixels are in RGBA space.