Character Recognition Using Graph-Based Method for Various Character Styles

M. Saravanakumar and S. Kannan

Abstract

Graph-based character recognition is a powerful technique that leverages the structural properties of characters by representing them as graphs, making it well-suited for recognizing characters with complex shapes and topologies. However, variations in handwriting styles and fonts pose significant challenges to the accuracy and reliability of these systems. This research investigates the robustness of graph-based character recognition to such variations, aiming to enhance its performance in real-world handwritten style variations using Attributed Relational Graphs (ARGs).

The study begins by analyzing how different handwriting styles and font variations affect the graph representation of characters, identifying key factors that contribute to recognition errors. To address these challenges, we develop novel graph construction techniques that normalize and standardize character graphs, reducing sensitivity to stylistic differences. Additionally, we propose adaptive graph matching algorithms that allow for flexibility in handling discrepancies caused by variations in style and handwriting.

The proposed methods are rigorously evaluated across diverse datasets, encompassing a wide range of handwriting styles, fonts, and noise levels. Our results demonstrate significant improvements in recognition accuracy and robustness, particularly in challenging scenarios with substantial variations in character appearance.

This research advances the state of the art in graph-based character recognition and provides valuable insights into the development of more resilient recognition systems that can generalize across different writing styles and fonts. The work has broad implications for applications such as digitizing handwritten documents, real-time handwriting recognition, and multilingual text processing.

Keywords

Attributed Relational Graphs (ARGs); Graph Edit Distance (GED); Spectral Matching; Subgraph Isomorphism; Graph Neural Networks (GNNs); Approximate Graph Matching; Harris and Shi-Tomasi Corner Detection

Corresponding Author

M. Saravanakumar, Department of Computer Science, School of Information Technology, Madurai Kamaraj University, India.

Citation

Saravanakumar, M., & Kannan, S. (2025). Character Recognition Using Graph-Based Method for Various Character Styles. Cognitive Computing and Extended Realities, 1(1), 01-14.