A Bengali Handwritten Vowels Recognition Scheme Based on the Detection of Structural Anatomy of the Characters

The present paper describes the development of a novel scheme for recognising Bengali handwritten vowels using mathematical morphology where the characters are first categorised on the basis of their anatomical features. The scanned images of the characters are passed through a weighted decision tree which is designed to analyse the feature set present in them. Based on the detected features, the algorithms easily recognise the individual characters. The extraction of some features has been carried out by generating the curvature scale spaces for the characters. The treatment is performed on scanned binary images of the handwritten characters to detect anatomical features such as 'bowl', 'lobe' and 'arm'. The scheme developed here is very fast and it does not need training samples to work.