Image processing based identity recognition and liveness analysis system with microservice

Dijital identity verification is a process used to identify and authenticate individuals or systems in digital environments. This process typically involves usernames, passwords, biometric data (such as fingerprints, facial recognition, retina scanning), one-time passwords, and other security protocols. Digital identity verification methods commonly used to enhance security in online accounts, financial transactions, health records, and other digital platforms aim to provide defense against unauthorized access while safeguarding user information. In this study, the goal was to extract card information from images of the Turkish Republic Identity Card using optical character recognition (OCR) methods. Additionally, facial similarity analysis and liveness analysis were conducted to determine if the person presenting the ID card is its legitimate owner. The CURL deep learning architecture was utilized for image enhancement, followed by YOLOv8 Instance Detection for card positioning with a 99.2% mAP score. A search algorithm was developed to detect the maximum area based on convex points, and this algorithm was optimized using genetic algorithms. The Turkish flag and facial image on the ID card were identified, and corner points of the card were determined through perspective correction. YOLOv8 object detection model for the Turkish flag achieved a 97.6% mAP score. The CRAFT model was used for text detection, followed by determining the optimal threshold value using the Otsu method. Image Super-Resolution (ISR) models were employed to enhance image quality, and text reading was performed using TesseractOCR, DonutOCR, and TrOCR models. For facial recognition and similarity analysis, the RetinaNet + Arcface model was used, while the 6DRepNet model with pose prediction architecture was utilized for liveness analysis. The system design was completed according to a microservices architecture, and the performance of the implemented system is presented in the study.