Hybrid Evolutionary Techniques to Restricted Feed Forward Neural Network with Distributed Error for Recognition of Handwritten Hindi “MATRAS‟
Meenu Gupta1, Ajay Rana2

1Meenu Gupta, Director, G L Bajaj Group of Institutions, Mathura (UP),
2Ajay Rana, Director, ASET, Amity University, Noida (UP).

Manuscript received on April 15, 2012. | Revised Manuscript received on April 20, 2012. | Manuscript published on May 05, 2012. | PP: 161-169 | Volume-2 Issue-2, May 2012 . | Retrieval Number: B0546042212/2012©BEIESP
Open Access | Ethics and Policies | Cite 
© The Authors. Published By: Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

Abstract: This paper evaluates the performance of restricted feed forward neural network trained by hybrid evolutionary algorithm with generalized delta learning rule for distributed error to obtain the pattern classification for the given training set of Handwritten Hindi ‘MATRAS’. Generally, the feed forward neural network considers the performance index as backpropagated instantaneous unknown error for output of hidden layers. Within this proposed endeavor, we are considering the performance index of distributed instantaneous unknown errors i.e. different errors for different layers. In this case, the convergence is obtained only when the minimum of every error on different layer is determined. The simulation for the performance evaluation is conducted for hand-written ‘MATRAS’ of Hindi language scripted by five different people. These samples are stored as scanned images. The MATLAB is used to determine the densities of these scanned images after partitioning each image into 16 portions. These 16 densities for each character are used as an input pattern of training set. We consider five trials for each learning method and results are presented with their mean value.

Keywords: Genetic Algorithm, Handwritten Hindi MATRAS, Multilayer Feed Forward Neural Network, Pattern Recognition