Enhancing Low-Resolution Facial Images for Forensic Identification Using ESRGAN

This research is motivated by the challenges in facial identification for forensic investigations due to poor image quality, especially from low-resolution CCTV recordings. Images with noise, low lighting, and suboptimal angles often hinder accurate facial recognition. This study aims to examine the effectiveness of the Enhanced Super-Resolution Generative Adversarial Network (ESRGAN) method in enhancing the quality of forensic facial images. The methodology consists of three main stages: data preparation of low-resolution facial images, applying the ESRGAN model to enhance image resolution, and evaluating the results using metrics such as PSNR and SSIM. The findings reveal that ESRGAN significantly improves the visual details of facial images, thereby supporting better facial identification processes. These results have important implications for leveraging deep learning technology to facilitate image analysis in forensic contexts. However, challenges such as extreme noise presence require further development of methods to achieve more optimal outcomes.