Writer Identification System Based on Offline Handwritten Text in Gurumukhi Script

Abstract

Writer Identification (WI) based on handwriting is an amazing application under the wide spectrum of machine learning and document based identification and recognition. Its most acknowledged applications include forgery detection, investigating forensic crimes and resolving other suspects. This paper is concerned with the development of WIS based on the handwritten text in Gurumukhi script. Through this paper, authors presented a contemporary and experimental move in Gurumukhi script, its framework, characteristics of handwriting modality, differences between writer identification and verification, text independent and text dependent systems, script dependent V s script independent, character set of Gurumukhi script, latest survey on WIS based on offline handwritten text in Gurumukhi (Punjabi) script and experimental findings. A dataset of 100 writers i.e. 100 x53 x 10=53000 Gurumukhi characters has been taken for the experimental study. Feature extraction techniques such as Zoning, Transition, Diagonal and Peak Extent based were used and for the classification part Multi Layered Perceptron (MLP), Artificial Neural Network (ANN), and Random Forest (RF) classifiers were implemented. The proposed experiment reported 93.06% writer identification accuracy rates along with performance evaluation metrics with 93.2% True Positive Rate (TPR) and 0.39% False Positive Rate (FPR). The reported result outperforms in comparison to the literature survey and also poses futuristic and upcoming directions to the researchers such as age, gender, handedness, physiological autopsy, personality, stress and even nationality identification based on handwriting.