An optimal clustering for fuzzy categorization of cursive handwritten text with weight learning in textual attributes

Abstract

A new method for fuzzy categorization of cursive handwritten text has been addressed in the present work. This is based on input text clustering and subsequent learning of weighted attributes in each subject cluster. The system first employs a new algorithm to detect the letter boundary in each cursive word in the textual sentences. A Modified Optimal Clustering Algorithm (MOCA) and Back Propagation (BP) Network combination converts the handwritten texts into printed ones. Subject wise grouping of printed texts are then made with Optimal Clustering Algorithm (OCA). The weighted attributes of each subject is thereafter learned to finally find out the fuzzy categorization of each input text. Different performance metrics of the system is computed with a newly introduced concept of Fuzzy Confusion Matrix. The performance evaluation of the fuzzy categorization of text with Holdout Method in terms of accuracy, precision, recall and f-score is appreciably high. Also, the learning and categorization time is quiet affordable.