Publication: Explainable artificial intelligence for signature verification system
Loading...
Date
2023-10
Authors
Lee, Sze Yuan
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
In recent years, the use of personal identity, such as signatures, as a means of authentication has gained significant attention. There are some concerns arise due to the potential for signature forgery and leading to the development of signature verification systems to determine the authenticity of signatures. The lack of
understanding behind the AI and DL can erode trust in the tools as incorrect or biased decisions made. The application of Explainable Artificial Intelligence (XAI) methods in signature verification systems can address these concerns by providing insights into the decision-making process and enhancing the trustworthiness and reliability of the system. This research aims to explore and evaluate various explanation models to improve the interpretability and performance of signature verification systems. Furthermore, this research seeks to identify the specific aspects that users and developers focus on when considering explanations generated by these models. Moreover, this research aims to develop a new explanation model by combining the strengths of two widely used methods, LIME and Grad-CAM. The experiment is conducted through MATLAB using package known as Deep Learning Toolbox. The explanation evaluates through the respond of 18 respondents in four aspects, understandability, interpretability, accuracy and usefulness. The survey is also used to identify the evaluation aspect that are focused by users and developers. In addition, a new explanation model is developed through the combination of “scoremap” of LIME and Grad-CAM. Preliminary findings indicate that the Grad-CAM method demonstrates better performance from the user's perspective, while developers tend to prefer the LIME method. By leveraging the strengths of both approaches, the new explanation model achieves an impressive increase in understandability,
interpretability, accuracy, and usefulness.