What is it about?

Face recognition systems can work very well on clear, high-quality photographs, but forensic investigators often have to work with images that are blurred, compressed, noisy, low-resolution, poorly lit, or captured from screens. These kinds of images can make automated face recognition much less reliable. This study investigates whether modern AI image enhancement can improve face recognition when images have been degraded in ways commonly seen in forensic and security settings. The research uses a latent diffusion-based enhancement pipeline to improve poor-quality face images before they are processed by face recognition systems. The method was tested across seven types of image degradation, including blur, compression, scaling, noise, colour distortion, motion blur, and screen recapture effects. Using 3,000 individuals and 48,000 recognition attempts, the study shows that AI-based enhancement can substantially improve recognition performance. Overall recognition accuracy increased from 29.1% on degraded images to 84.5% after enhancement. The results were also checked using biometric evaluation measures, confirming that the improvement was not simply due to changing a decision threshold.

Featured Image

Why is it important?

This work is important because real forensic evidence is often far from ideal. Surveillance footage, mobile-phone images, social media images, and screen-captured photographs may be too degraded for reliable face recognition. Improving the quality of such images before recognition could make existing forensic face recognition systems more useful in practice. What is distinctive about this study is that it evaluates enhancement under realistic forensic degradation conditions and compares the proposed approach with established image restoration methods. The results show that the diffusion-based pipeline outperformed ESRGAN, Real-ESRGAN, and CodeFormer, achieving substantially higher recognition accuracy. The study also tested whether the benefits held across different face recognition backbones, helping to show that the improvements reflect genuine image quality enhancement rather than a result specific to one recognition model. The work could help forensic laboratories, law enforcement agencies, and security analysts make better use of low-quality imagery. At the same time, it highlights the need for careful governance: enhanced images should support expert analysis, not replace the original evidence, and their use must be accompanied by transparency, provenance records, and ethical safeguards.

Perspectives

This research shows that image enhancement should be judged not only by whether an image looks clearer, but by whether it improves recognition performance in a measurable and forensically meaningful way. The findings suggest that latent diffusion models can play a valuable role as a preprocessing step for forensic face recognition, especially when images are affected by blur, noise, or resolution loss. The study also points to the limits and responsibilities of this technology. Some degradations, such as repeated JPEG compression, remain more difficult to recover from. In forensic settings, enhancement must not be treated as creating new evidence or replacing human expertise. The original image must be preserved, the enhancement process must be documented, and any use of enhanced images should be subject to legal, ethical, and fairness review. This work, therefore, supports a future in which AI enhancement tools help forensic practitioners extract more reliable information from poor-quality images, while maintaining transparency, accountability, and respect for evidential integrity.

Professor Hassan Ugail
University of Bradford

Read the Original

This page is a summary of: Evaluation of latent diffusion enhanced face recognition under forensic image degradations, Discover Computing, April 2026, Springer Science + Business Media,
DOI: 10.1007/s10791-026-10082-4.
You can read the full text:

Read

Contributors

The following have contributed to this page