Ashlin Deepa R.N¹, R. Rajeswara Rao² and Y. Vijayalata¹, ¹Gokaraju Rangaraju Institute of Engineering and Technology, India and ²Mahatma Gandhi Institute of Technology, India

Image pattern matching is one of the most widely used techniques in character recognition. A pattern is the description of an object. Although a large number of papers have been reported on handwritten character recognition, it is still a very challenging problem. Recognition of characters is based on pattern classification. In this paper a hybrid approach is proposed using fuzzy technique and multiset comparison to recognize handwritten characters. Fuzzy technique is used to measure the variations in the features of handwritten numerals to form strings of cha-racter. Features are selected from the preprocessed image and the feature is labeled using fuzzy logic. Strings of a character are formed from these labeled primitives. Each string is divided in-to a multiset of puzzle pieces by repeatedly applying a mask which is comprised of all ones. The string is padded with anchors at the beginning and end. The mask is placed over the string, be-ginning at a certain character, and reading off all characters corresponding to 1 bit in the mask, thus producing one puzzle piece. To divide a string into multiset of puzzle pieces, the mask is applied at all shifts in the string starting at each character. Thus for each prototype character, we associate a multiset of puzzle pieces, which form the database. The multiset of puzzle pieces of an unknown character is constructed and compared with the same in the data-base, leading to the identification of the unknown character.

Fuzzy technique; Thinning; Segmentation; Labeling; Multiset; Puzzle pieces