AI-POWERED IMAGE ENHANCEMENT IN FORENSIC APPLICATIONS: CHALLENGES AND OPPORTUNITIES

Abstract

Recent progress in machine learning (ML) and computer vision has markedly enhanced the performance of automated facial recognition systems. Nonetheless, forensic applications frequently face issues with poor-quality and low-resolution images that pose challenges even to the most advanced facial recognition technologies. This research explores the potential of neural-based image enhancement and restoration techniques to recover degraded images while maintaining original correspondences for legal use. We evaluate super-resolution and deconvolution methods across two comprehensive and varied facial datasets. Our study employs twelve distinct GAN- and diffusion-based enhancement techniques. Our results offer insights and recommendations for the effective application of image enhancement in forensic facial recognition, highlighting the advantages and potential limitations of these advanced technologies.