From Shadows to Faces: Retinex–GAN and TransYOLO-FaceLite for Tiny and Occluded Face Detection in Low-Light CCTV

Read full text Summarise with VIRAT
Authors: Sony Priya S, Minu R I
Cited by: 0
References: 17

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Detecting tiny and occluded faces in low-light CCTV footage remains challenging due to severe illumination degradation, sensor noise, low spatial resolution, and long camera-subject distances.This paper presents a detection-driven, end-to-end face detection pipeline that integrates Retinex-GAN-based illumination enhancement, illumination-adaptive denoising, selective face-region super-resolution, a lightweight transformer-augmented YOLO detector (TransYOLO-FaceLite), and spatial-consistency refinement.Unlike perceptual enhancement approaches, the proposed framework jointly optimises enhancement and detection to preserve facial micro-structures critical for small and occluded faces.Experiments conducted on multiple low-light CCTV datasets demonstrate that the proposed method achieves a VOC-style mAP@0.5 of 86.7%, outperforming YOLOv5-Face, RetinaFace, DSFD, and TinyFace by 4.3-9.1%,with particularly strong gains on tiny-face and heavy-occlusion subsets.Ablation studies confirm consistent incremental improvements from each module, while maintaining real-time performance suitable for practical CCTV deployment.

Fields of Study & Subjects

Face recognition and analysis Advanced Neural Network Applications Face and Expression Recognition

Publication details

Journal / source
International journal of intelligent engineering and systems
Volume
19
Issue
4
Pages
717–736
Publication date
23 March 2026
Language
EN