Vol. 7, Special Issue 2 (2025)

This review paper comprehensively examines the evolution and current state of criminal face recognition technologies driven by artificial intelligence and deep learning. It traces the transition from traditional image processing techniques to modern convolutional neural network (CNN) architectures that have significantly enhanced the accuracy and efficiency of facial recognition systems. We explore various methodologies such as Siamese networks and triplet loss frameworks are critically analysed for their effectiveness in learning robust face embeddings, which are crucial for distinguishing individuals under diverse conditions including varied lighting, occlusions, and non-frontal poses in this investigation. The integration of these advanced techniques into law enforcement applications ranging from real-time surveillance and database matching to post-incident forensic investigations is explored in detail. Furthermore, the review discusses inherent challenges such as dataset biases, adversarial vulnerabilities, and the ethical and privacy implications of deploying such technologies. The potential benefits of multi-modal biometric integration are also considered as a promising avenue for improving system resilience and accuracy. This paper identifies gaps in current research and outlines future directions aimed at developing more robust, fair, and ethically responsible criminal face recognition systems.