Project Title: Bengali hand-writing recognition
Project Leader and Supervisor: Zahidur R Chowdhury
Other Member(s): Mohammad Abu Naser [Undergrad Thesis Student] and Ashraf Bin Islam [BSc Engineer]
My role: Supervise team members, establish initial frame work of the project, plan and keep team members on the schedule, and report writing.
Tools used: MATLAB (neural network) / LATEX / Microsoft Office
Summary
- Bengali hand-writing recognition has potential application in document processing for one of the widely used language in the world. A method using Artificial Neural Network (ANN) is utilized primarily to identify numerals of the language using transition features. Maximum accuracy of 82% was reported for the optimized network. The typical performance of the handwriting recognition system that uses a single recognition scheme is around 85%. Incorporation of the significant local features in a character was recommended to enhance the overall performance of the network based on the study.
- Figure: (left) Performance of the numeral recognition system (right) illustration of the transitional features in numerals that was used in the recognition process.
Publication(s)
Zahidur R. Chowdhury, M. A. Naser and Ashraf Bin Islam, “Bengali handwritten numeral recognition using artificial neural network and transition elements”, J. of Institute of Enggineers, Bangladesh, vol. 31, pp. 47-53, 2006. view / link