Understanding Urdu Ligatures

Abstract

In order to recognise Urdu ligatures in the Nastaliq typeface, an efficient segmentation-free and scale-invariant algorithm is presented in this study. The suggested method is based on training distinct hidden Markov models for the primary body of ligatures and the ancillary components. Using right-to-left sliding windows, features capturing the projection, concavity, and curvature information of ligatures are retrieved and supplied to the models for training. The system achieved a recognition rate of 97.93% after being trained and tested on a total of more than 2,000 frequently occurring Urdu ligatures from a standard database.