DEVELOPMENT AND PERFORMANCE ANALYSIS OF SIGNATURE AUTHENTICATION SYSTEM LEVERAGING SPARSE REPRESENTATION FOR FEATURE EXTRACTION

• Main Author: Thalca Avila Putra, Aira
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85249
• Institution: Institut Teknologi Bandung

Biometric technology has become a cornerstone in various identification and authentication systems, with signature being one of the most widely used methods. Currently, traditional authentication methods often lack the accuracy of those that leverage deep learning. However, deep learning-based authentication models often require significant computational resources and large model sizes. This research aims to address these challenges by developing a framework that integrates sparse representation into a machine learning model for signature authentication. The proposed framework is designed to produce a lightweight, fast, and accurate model. The framework was tested on the CEDAR signature dataset, achieving a balanced accuracy of 98,52% for skilled forgeries and 97,99% for random forgeries, with an inference time of less than 0.2 seconds and a model size under 500KB. The results demonstrate that the developed system is not only accurate but also efficient in resource usage. This research contributes significantly to improving the efficiency of signature authentication systems and open up new opportunities for applying sparse representation methods to various machine learning models.