I've played with TinEye a bit, and never had much luck understanding the circumstances under which it works well or poorly. I don't think it's ever found anything of mine on APUG, but it clearly crawls Flickr---yet it often fails to recognize my stuff there, as well. But when it does find a match, it's often notably successful at recognizing different treatments of the same image---different toning, sharpening, cropping, stuff like that.

Google seems to have a worse matching algorithm but a larger crawl database---it found some instances of my images being linked from sites that aggregate Flickr content, though interestingly it didn't find the originals at Flickr itself---nor, again, at APUG. (Is the gallery protected by a robots.txt file, maybe? That would explain a lot.)

In the defense of both, soft matching of image or audio content is INCREDIBLY hard. We don't understand squat about how our brains do signal processing, and transformations that hardly even register with a human observer can completely flummox a quantitative algorithm. As a professional in a related field, I'm amazed that TinEye works as well as it does and I'd love to know what goes on under the hood.

-NT