A learning based handwritten text categorization

In the present work a learning based handwritten text categorization technique using Malsburg Learning BP Network combination has been designed and developed. This combination is used to identify alpha numerals and thereafter convert the handwritten text into printed one via appropriate word formation. Groups of text belonging to different subjects are fed to this system, and the system extracts the salient features in terms of attributes in intra groups. The commonality among the inter groups are thereafter discarded. The attributes or salient features thereby learned are later used as glossary for each group for performing unlabeled text categorization. The performance evaluation of this system with labeled test texts using standard Holdout Method in terms of accuracy, precision, recall, f-score is appreciable. Also the learning and performance evaluation time is affordable.