Bi-lingual Handwritten Character and Numeral Recognition using Multi-Dimensional Recurrent Neural Networks (MDRNN)

Abstract

This work describes the implementation of a multidimensional RNN and a MLP- RNN hybrid ANN structure for developing models of character recognition applied to AngloAssamese scripts and numerals. The first stage of the work is related to the feature extraction process of characters and numerals of Assamese- an important language in the North Eastern part of India. The second stage of the work is related to the implementation of multiple ANNs in cooperative format for handling Assamese scripts and numerals. The third part of the system deals with Anglo- Assamese scripts and keeps track of performance of ANNs while tackling such mixed inputs implemented in cooperative systems for ascertaining performance levels while handling mixed, time and size varying inputs. A detailed description is also included regarding the experiments performed and the results derived. Finally, a brief account concludes the description and provides likely directions the work can take in the future.