Investigating the Efficacy of Forensic Facial Reconstructions: A Dual Approach Using Sketches and CCTV Images
Pranav V Jambur, Smruthi S Kadagadkai · 2024
The evolution of forensic facial recognition has driven the development of a groundbreaking two-model system engineered to identify suspects by digitally removing facial obstructions such as masks and glasses. Throughout history, CCTV captures and forensic sketches have been pivotal in criminal investigations, yet their efficacy has been limited by challenges like low resolution and poor lighting conditions. This innovative system harnesses Generative Adversarial Networks GAN to elevate inputs from CCTV captures and forensic sketches. Model A excels at transforming forensic sketches into realistic images, refining resolution and color fidelity to enhance accuracy in suspect identification. Model B integrates GFP-GAN, pix2pix GAN, and StyleGAN to effectively eliminate facial obstructions, thereby refining outputs for heightened realism. Leveraging advanced image processing techniques, including convolutional neural networks and transfer learning, Model B adeptly handles diverse and challenging conditions. Methodologically, the system involves meticulous dataset collection, rigorous training, and precise processing utilizing the CelebAMask-HQ dataset for Model A, and seamless integration of GFP-GAN, pix2pix GAN, and StyleGAN for Model B. Demonstrated results underscore the system's prowess in processing varied inputs and bolstering facial recognition accuracy, pivotal for applications in law enforcement and national security. Authorities benefit from expedited identification processes, facilitating swift suspect tracking and crime prevention. Looking ahead, ongoing advancements aim to integrate Neural Radiance Fields into the GAN framework, promising enhanced visual fidelity and enabling real-time suspect identification. This continuous evolution marks a paradigm shift in forensic facial recognition capabilities, empowering law enforcement agencies worldwide with robust tools for effective crime mitigation.