I have seen few reverse image search engines such as Google Lens, SauceNAO, IQDB and etc.
And I wonder how these engines work with very fast speed, even they have to load image from other site(In case of sauceNAO, they get illustration works from pixiv - which has about 1 million arts in total) and check it's image similarity.
Can anyone explain me how these engines work?
They use image fingerprinting/hashes, in image hashes similar input results in similar output not like crypthgraphic hashing where one small change in the input produces a completely different output.
So if they have indexed a lot of image hashes that they find by crawling the web, then they can quickly check if the hash of the uploaded image is similar to one that they have indexed