AFAIK it compares thumbnails, where the text wouldn't make much difference.
It shows the thumbnail, but probably checks the CRC code or other hash. Although that means invisible change in the file (such as tags in the file) might cause it to slip the system. Or that issue about Pixiv "optimizing" (stripping) the files making it smaller than version uploaded elsewhere.
Anyway, is this from one of those many Facebook pages ? If yes, they probably don't care much about font or even if they let some translation mistakes slips through. It will be unreachable, buried under pile of "new posts" in few hours anyway. They just need it to get more Likes to boost their page's marketability.
It shows the thumbnail, but probably checks the CRC code or other hash. Although that means invisible change in the file (such as tags in the file) might cause it to slip the system. Or that issue about Pixiv "optimizing" (stripping) the files making it smaller than version uploaded elsewhere.
If that were the case it'd be "find exact copy" not "find similar." I haven't found piespy mentioning exactly how iqdb works, but forum grazing says it apparently it involves comparing Haar wavelet analysis of the images. Hints were dropped that it compares the JPEG thumbnails instead of entire images, but that was like 7 years ago.
Then again, I don't know what similarity threshold level has to be met for it to appear on the new single-example given on the upload page, compared to when it used to take you to an external page that gave a wall of possible images.
