Psychological image similarity algorithm / library

I am looking for an algorithm (ideally a C / C ++ implementation) that calculates the perceived similarity between two images, taking into account psycho-visual factors (for example, that the difference in color is not as bad as the difference in brightness).

I have the original image and several variations of it (256-color quantizations in my case), and I would like the algorithm to find which image the person will evaluate as the best.

The best I have found so far is SSIM , but it does not "understand" anti-aliasing (error propagation) and the implementation uses linear RGB (I fixed that it implements my own ).

Alternatively, it may be an algorithm that preprocesses the images for comparison with SSIM / PSNR / MSE or another typical algorithm.

+3
source share
1 answer

Well. Can you turn it into an online job with an Amazonian mechanical Turk? Or make a game like google image labeler ? You can give extra points or pay if people agree with their ratings.

, . SSIM , , SSIM, , . , , .

0

Source: https://habr.com/ru/post/1777330/


All Articles