Hacker News Reader

OP's method relies on the permutation of elements in (columns or rows) to be in the same order for each (column or row).

If you're just handed a pile of confetti you don't have that same correspondence.

View on HN

Also, you generally have written correspondence, not images. I doubt that the similarity metric of OPs method produces reasonable results on text, and a good similarity metric for text would be much more complicated.

On the other hand physical shredders typically have other weaknesses, for example if you don't stir the shredded result you have a high correlation between original the current and original relative position of the individual pieces.

With written text, you would expect that at least some the letters from one shred would continue on its neighbors. I don't see why the similarity metric would not work. Also normally you have a pretty high contrast between the letter and the background, unlike the image that was used in the example.

If you have a language model, then sequences of characters appearing in your text will add further information to your edge matches.

Reply on news.ycombinator.com