Journal of Pattern Recognition Research

Devnagari script is the major script of India and is widely used for various languages. In this work we propose a new shape based technique for recognition of isolated handwritten Devnagari characters. The character set consists of 33 consonants and 11 vowels. The character is segmented in two parts when the presence of top modifier is detected above the topline using Hough Transform. The character is then segmented in various segments (strokes) using basic structural features like endpoint, crosspoint and junction points. We propose modified direction codes for segmented strokes based on our thinning algorithm to first classify the strokes as left curve, right curve, horizontal stroke, vertical stroke and slanted lines etc. The knowledge of script grammar is applied to identify the character using shapes of strokes, mean, relative strength, straightness and circularity and their location based on fuzzy classification. Characters are first pre-classified using a tree classifier and then unordered stroke classification is used for final classification. The system tolerates slant of about 10º left and right and a skew of 5º up and down. The average accuracy of recognition varies is 86.4%.