Algorithm to find edit distance to all substrings

To find substrings in a given string is very easy. You take the normal Levenshtein algorithm and modify it slightly.

FIRST: Instead of filling the first row of the matrix with 0,1,2,3,4,5,... you fill it entirely with zeros. (green rectangle)

SECOND: Then you run the algorithm.

THIRD: Instead of returning the last cell of the last row you search for the smallest value in the last row and return it. (red rectangle)

Example: needle: "aba", haystack: "c abba c" --> result = 1 (converting abba -> aba)

I tested it and it works.

This is much faster than your suggestion of stepping character by character through the string as you do in your question. You only create the matrix once.

The Wagner-Fischer algorithm gives you the answer for all prefixes "for free".

http://en.wikipedia.org/wiki/Wagner%E2%80%93Fischer_algorithm

The last row of the Wagner-Fischer matrix contains the edit distance from each prefix of s to t.

So as a first crack at your problem, for each i, run Wagner-Fischer and select the smallest element in the last row.

I will be curious to see if anyone else knows (or can find) a better approach.