A Hybrid Recognition Scheme Based on Partially Labeled SOM and MLP

We propose a hybrid system for Bangla handwritten numeral recognition based on partially labeled two-layer SOM and MLP classifiers. Partially labeled mechanism is introduced to the Kohonen's SOM for reducing recognition error rate, and two-layer structure is applied for improving the performanc...

Full description

Saved in:
Bibliographic Details
Main Authors: Shujing Lu, Chunyun Xiao, Yue Lu
Format: Conference Proceeding
Language:English
Subjects:
Character recognition
Computer science
Error analysis
Feature extraction
Feeds
Handwriting recognition
Multilayer perceptrons
Natural languages
Neurons
Sorting
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:We propose a hybrid system for Bangla handwritten numeral recognition based on partially labeled two-layer SOM and MLP classifiers. Partially labeled mechanism is introduced to the Kohonen's SOM for reducing recognition error rate, and two-layer structure is applied for improving the performance of the SOM classifier. The directional and density features are utilized in our system, and the partially labeled SOM is applied first. In the case that the character cannot be recognized by the partially labeled SOM, it will be feed to a multi-layer perceptron classifier for further processing. The experiments on the Bangla handwritten numeral samples captured from real envelopes have found that the hybrid system achieves 96.7% correct recognition rate
ISSN:1051-4651
2831-7475
DOI:10.1109/ICPR.2006.85