A model of a feed forward multi layered neural network to recognize hand written Bengali digits

Research output: Contribution to conferenceConference paper

Abstract

This work deals with the recognition of handwritten Bengali digits using feed forward multi layered neural network approach. It proposes an overall guideline of how to construct and train a neural network in order to give it the recognition capability. The paper presents findings of different experiments undertaken and concludes on the quality of recognition. The key aspect of this research is the solution network that successfully recognized handwritten digits with as high accuracy as 96% on casually written digits and 100% accuracy on carefully written digits. It also features two very simple, yet interesting and efficient network structures that were used. The advantage of these features is the small size, which takes smaller storage space and helps to train faster. Another major achievement is the implementation of a new activation function which theoretically reaches 'zero error' at some stage of the training. A sophisticated conversion and compression algorithm for digitizing the handwritten images has also been developed and described in the paper.
Original languageEnglish
Pages331-338
Number of pages8
Publication statusPublished - 1 Dec 2010
Externally publishedYes
EventProceedings of the 2010 International Conference on Artificial Intelligence, ICAI 2010 -
Duration: 1 Dec 2010 → …

Conference

ConferenceProceedings of the 2010 International Conference on Artificial Intelligence, ICAI 2010
Period1/12/10 → …

Keywords

  • Digits
  • Handwritten
  • Multi layered
  • Neural network
  • Training

Fingerprint

Dive into the research topics of 'A model of a feed forward multi layered neural network to recognize hand written Bengali digits'. Together they form a unique fingerprint.

Cite this