AI-Driven Passwordless Authentication through MobileNetV2-Based Dynamic Password IrisCode Generation and Deep Iris Embedding Transformation

Abstract

The widespread adoption of digital services, cloud infrastructure and smart access control has raised the demand for passwordless and secure authentication systems. The classic password-based authentication is susceptible to stolen credentials, phishing and replay attacks, leading to unauthorized access, whereas traditional iris authentication methods rely on fixed biometric templates with less flexibility and security. Here, proposing an AI-based passwordless authentication system using dynamic generation of Password IrisCode and iris embedding transformation. This framework utilizes a custom-designed Iris Acquisition Device with controlled lighting and a high-resolution capturing facility to secure a good-quality iris image. This captured data is then pre-processed and fed to the MobileNetV2 model to extract distinct iris features and generate compressed 128-dimension biometric features (iris embedding). Instead of traditional iris authentication using an iris feature vector, this extracted embedding is cryptographically hashed (using SHA-256) and converted to an encrypted Password IrisCode, thus securing the data while retaining the uniqueness. The proposed system authenticates the users by checking the similarity of generated iris features and the enrolled patterns. The framework is tested on a large dataset with variable illumination and acquisition conditions called CASIA-Iris-Thousand and has been able to achieve an accuracy of 95.0%, as against the accuracy of 67.09% of the IrisCode authentication system, and a failure rate has been reduced from 32.91% to 5.0%, and average processing time has been observed to be 0.4065sec for near real-time authentication. The AI-based feature extraction and the usage of dynamic Password IrisCode Generation, along with a dedicated hardware for iris acquisition, make the proposed system efficient, secure and scalable for modern passwordless access controls.