An Analytic Scheme for Online Handwritten Bangla Cursive Word Recognition

In this article, we describe a prototype system for recognition of online handwritten cursive words of Bangla, a script used by more than 200 million people of India and Bangladesh, two neighboring countries of Asia. To the best of our knowledge, in the literature, there does not exist any work on recognition of such online Bangla cursive words. Here, we propose an analytic recognition approach, which involves segmentation of the input Bangla word. Modified quadratic discriminant function classifier is used for recognition of segmented strokes based on a chain code histogram based feature vector. Finally, an input word is recognized by a verification module, which uses a set of rules for construction of characters from strokes. We carefully selected a set of 100 Bangla words such that each basic character and vowel modifier of this script occurs at least in two words. A total of 10000 handwritten online word samples provided by 50 native Bengali writers of different groups with respect to age, education, sex and income have been used in the present study.