ElectroSphere: Electrical Electronics Engineering Bulletin
Resolving Node Identification Issues in Graph Based Method for Character Recognition
Abstract
M. Saravana Kumar and S. Kannan
Character recognition has evolved as a critical component in various researches such as image processing, automated data entry, and digital archiving. Traditional methods often struggle with variations in font, size, and orientation, making robust recognition a challenging task. This paper proposes a novel graph-based method for character recognition, which leverages the structural and topological properties of characters. In the proposed approach, characters are represented as graphs where nodes correspond to critical points (e.g., junctions, endpoints) and edges represent the connecting strokes. This graph representation preserves the geometric and structural information of the characters, enabling more effective handling of variations. The recognition process involves the following key steps:
• Graph Construction: Extracting significant points and constructing the graph representation of the character.
• Feature Extraction: Utilizing graph-theoretic features such as node degree, path length, and sub graph isomorphism to capture the unique characteristics of each character.
• Graph Matching: Comparing the constructed graph with pre-defined template graphs using graph matching algorithms to identify the character.
• Classification: machine learning techniques to classify the character based on the extracted features and matching results. The proposed method is evaluated on standard character recognition datasets, demonstrating superior performance in terms of accuracy and robustness against distortions and variations compared to traditional pixel- based methods. This graph-based approach provides a promising direction for future research and applications in character recognition.