Enhancing Signature Authenticity Recognition Using CNN, PCA, and DAG: A Study on Accuracy Improvement and Parameter Refinement

This study focuses on enhancing signature authenticity recognition by leveraging Convolutional Neural Networks (CNN) combined with Principal Component Analysis (PCA) and Directed Acyclic Graphs (DAG). While CNNs have demonstrated promising results, discrepancies between training and test accuracies remain a challenge. By integrating PCA for dimensionality reduction and DAG for optimizing training, this research aims to improve efficiency and accuracy. Despite achieving $\mathbf{1 0 0 \%}$ training accuracy, the test accuracy reached only $\mathbf{9 7 \%}$, indicating the necessity for refining parameters further. This study emphasizes the value of systematic methodologies and comprehensive parameter exploration to achieve substantial advancements in signature recognition accuracy.