Hacker News Reader

This looks trivial to defeat by simply comparing the top and bottom edges of each image, and picking the combination with the best fit. The faces might fare a little better under that test, but the logos and colourful shapes would likely be successfully matched every single time.

View on HN

Indeed. This is absolutely trivial to defeat, and anyway the space of possibilities is just too narrow. Even with faces, the amount of similarity between top pixels of the middle row, and bottom pixels of the top row, are a lot bigger when there is a match.

Actually, logos are the harder of the pair. My simple attempt only gets around 90% success because some of the logo pieces have a group of white at the top and bottom. This makes it so the bottom piece of a logo and the top piece of the logo make good fits for the middle and bottom rows. Example from my solver: http://imgur.com/CQvcH.png

The photos take up the entire frame with various colors instead of one solid one, so it is not likely for any of them to end up with a mismatch like this.

Oh, that's a really interesting case! I hadn't seen any like that when I posted, but that's definitely something to consider. I think you could work around whenever you get multiple "perfect" fits like that by comparing a) whether they are non-blank, or, more sophisticated, b) are there continuous edges across the boundary.

Reply on news.ycombinator.com