Performance Evaluation of Transfer Learning Techniques and Machine Learning Methods in Signature Verification

Signature verification plays an important role in biometric security systems and traditional methods can lead to limitations in verification accuracy. However, traditional signature verification techniques work with limited data and features, which can negatively affect the accuracy of verification processes. In this study, we investigate the performance improvement in signature verification by combining transfer learning and machine learning algorithms. In the experiments performed on signatures from the BHSig260 Hindi dataset, the transfer learning models (ResNet50, MobileNetV2, VGG16, InceptionV3, EfficientB7, DenseNet169) achieved high accuracy rates on their own, especially the VGG16 model performed the best with 93.77% accuracy. In the later stages of the study, machine learning algorithms such as K-nearest neighbor (KNN), Support Vector Machines (SVM) and Random Forest were added to the transfer learning models to further improve the validation performance. The combination of EfficientB7 + Random Forest achieved the highest performance with 95.24% accuracy. The results show that the integration of transfer learning techniques with machine learning algorithms significantly improves the accuracy of signature verification tasks. This combination stands out as an effective method that significantly improves the reliability and efficiency of biometric security systems. The findings of the study will provide an important reference for the future development of signature verification systems, contributing to the development of more accurate and reliable solutions in this field.