Real-Time Pashto Handwritten Character Recognition Using Salient Geometric and Spectral Features

Abstract

Pashto scripts are cursive in nature and hard to recognize in real-time. Native speakers of the Pashto language are large in numbers and reside in different regions of the world. Due to the cursive nature of the Pashto script along with variations in character strokes, the printed, as well as handwritten characters, are difficult to be detected, classified or recognized. In real-time handwritten character recognition systems, the challenging factors that constraints the system depends on the stroke noise, geometric behavior (like rotation, scaling and shifting, etc.) of the text. In this article, we provide an efficient technique that aimed to recognize handwritten characters in real-time by first smoothing the noise components in the text and then extract shape-based invariant features from the handwritten strokes. For real-time recognition of characters, the probability-based multi-class Naïve Bayesian classifier is exploited, which determines the probabilities of geometric invariant features to predict the character with the highest likelihood. The performance of the proposed approach has been validated through extensive experiments and based on the recognition matrices, the proposed technique achieves an accuracy of 97.5% for online Pashto handwritten characters in real-time.