Online Handwriting Identification from Khmer Script Using a Deep Learning Approach with KhNet

Abstract

Identifying a writer through handwriting is a complex yet essential task in pattern recognition, especially for script-based authentication. This paper presents a novel approach based on a Siamese neural network (SNN) for Khmer handwriting writer identification. The SNN architecture was applied to both the training and testing phases using a specially curated to ensure accuracy, consistency, and relevance for Khmer writer identification. The curation process involved several key steps: cleaning, to remove incomplete, noisy, or corrupted handwriting samples; validating, to ensure that all image pairs were correctly labeled as either positive (same writer) or negative (different writers); and enhancing, which included resizing images, converting dynamic handwriting traces into grayscale renderings, and improving contrast to highlight subtle characteristics of Khmer script. Extensive experimentation demonstrated the ability of the model to perform a high rate of accuracy on both the training and testing sets. KhNet, our proposed SNN method, effectively addressed the complexities of Khmer handwriting, accurately identifying the authorship of words. This research contributes to the advancement of writer identification techniques for Khmer script, with potential applications in forensic analysis, document verification, and linguistic studies. Future work will focus on enhancing the robustness, scalability, and incorporating sequential features such as (x, y) coordinates to further improve accuracy.