A Graph-Based and Pattern Classification Approach for Kannada Handwritten Text Recognition Under Struck-Out Conditions

This research focuses on the processing and identification of handwritten Kannada text, particularly under struck-out conditions. The database considered in this study comprises handwritten data. When such a database is processed using optical character recognition (OCR)-based digital systems, the output may often be in an unrecognizable format. To address this issue, a model has been developed incorporating pattern classification and a graph-based method for text identification. For pattern classification, feature extraction is performed using two different classes with a support vector machines (SVMs) classifier. In the graph-based approach, struck-out strokes are analyzed using the shortest path algorithm. To handle zigzag or wavy struck-out Kannada text, all possible paths of the strike-out strokes are identified, and suitable features are extracted for further processing. The synthesized/recovered text is processed using an inpainting cleaning method to ensure text recovery. The proposed methodology has been tested on both trained and untrained datasets of Kannada script. Performance evaluation was conducted using three parameters: precision, F1 score, and accuracy.