Kar sila stirmali grenme ile Tek-Ati s Imza Tanimlama

Handwritten signatures are a powerful biometric authentication tool that reflects an individual's unique writing style. Despite the widespread digitization, they continue to maintain their importance in critical areas such as identity verification, data integrity, and legal validity. Although existing signature recognition methods successfully determine whether a signature is real or fake, they face some limitations in signature identification, which is a multi-class classification problem. One of the main limitations is the need to retrain the model when a new user is added to the system. In this study, a signature identification method is proposed that eliminates this limitation through a contrastive learning approach. The model developed using supervised contrastive learning ensures the continuity of the system without requiring retraining when new signatures are added, thanks to the one-shot technique. Experimental results show that the proposed method is 20% more successful than base model.