Global geometry extraction for fuzzy logic based handwritten character recognition

The problem of handwritten character recognition is made more difficult by the fact that scanned, isolated characters are rotated or slanted in either direction and by a variable amount. This paper provides a possible solution to the problem of character rotation in the offline recognition of isolated handwritten English alphabets. The proposal involves extracting rotationally invariant global geometric properties from a character's image to analyse its shape. The geometric properties extracted are added to the input vectors of two existing classifiers. Experimental results tested on a variety of writing styles have shown that the inclusion of global geometric property extraction can improve the recognition rate by about 40%.