A COMBINED MACHINE LEARNING APPROACH FOR IMAGE BASED BANK CHEQUE SIGNATURE VERIFICATION

Abstract

Signature recognition and verification is a crucial task in the field of biometrics, which aims to identify individuals based on their unique physiological or behavioral characteristics. In recent years, machine learning algorithms have been widely used for signature recognition and verification due to their high accuracy and efficiency. One such algorithm is the K-Nearest Neighbors (KNN) algorithm, which is a non-parametric method used for classification and regression tasks. However, KNN has limitations in handling complex data structures such as graphs. To overcome this limitation, Graph Neural Networks (GNNs) have been proposed as an effective solution for graph-based data. In this research, we propose a signature verification model by combining KNN with GNN algorithms. The proposed model first extracts features from the signature image using Freeman Chain Code (FCC). We used a signature Database called CEDAR Signature dataset which consists of 792 signatures. These features are then used to train the KNN classifier, which is responsible for identifying the nearest neighbors of a given signature. However, instead of using the traditional Euclidean distance metric, we use a graph-based distance metric that takes into account the structural information of the signature. To further improve the performance of the system, we incorporate GNNs into the KNN classifier. The GNNs are used to learn the underlying graph structure of the signatures and capture their local and global dependencies. This allows the model to handle complex signatures with varying shapes and sizes. We use 82 percent of the data set for training and the remaining for testing. The experimental results show that our proposed model achieves 91 % accuracy and outperforms state-of-the-art methods. One major constraint/weakness of the study is the potential sensitivity to variations in signature styles and dynamic aspects that are not fully captured by the proposed model. To address this limitation, future research could explore incorporating additional dynamic features such as the velocity and acceleration of the signature strokes. Additionally, considering temporal information and capturing the sequence of strokes in a signature might enhance the model's ability to handle variations in writing styles.