Performance Analysis of Zone Based Features for Online Handwritten Gurmukhi Script Recognition using Support Vector Machine

Abstract

Handwriting recognition is a technique that convert handwritten characters into machine processable formats. The process of converting scanned images of handwritten text converted to machine understandable format is termed as offline handwritten character recognition, whereas online handwritten character recognition is the process of converting pen-tip movements (in the form of strokes) to machine understandable text. A good amount of research is going in this area since last thirty years. Various features for recognition of handwritten text have been proposed since its origin. Zone based features have widely been used for offline recognition of handwritten text. In this work, zone based features have been implemented on an online handwritten data collected from 30 users for a set of 82 unique middle zone strokes of Gurmukhi script. Support vector machine with various kernels and parameters like learning rate (ε), tolerance limit of termination (γ) and number of folds (k) have been explored for optimal performance. Zone based feature when applied with SVM gave 92.09% accuracy for diagonal features extracted from online handwritten text.