Skip to content
Conference Open access

NOVA-Face: CLAHE Enhanced YOLOv8 for Face Detection in Low-Light Surveillance

Oct 2026 · International Conference on Communication and Electronics Systems · pp. 1687-1692 · 0 citations · 33 references

Abstract

Face detectors such as RetinaFace and SCRFD perform well in well-lit scenes but degrade sharply in low-light surveillance footage, where sensor noise, motion blur, uneven illumination and very small faces are common. This paper presents NOVA-Face, a two-stage pipeline for low-light face detection. First, images are brightened with gamma correction and converted to LAB color space, where Contrast Limited Adaptive Histogram Equalization (CLAHE) is applied to the luminance channel to recover local contrast without altering color. Second, a COCO-pretrained YOLOv8n detector is fine-tuned on the enhanced images, first at 640 px and then at 1280 px. The model is trained on a unified dataset of 33,166 images (241,936 faces) drawn from DARK FACE, WIDER FACE and LLVIP, and evaluated on a validation set of 4,426 images (49,987 faces). On this validation set NOVA-Face reaches 73.2% mAP50 and 39.1% mAP50–95; on the night-time DARK FACE images alone, which are the focus of this work, it reaches 63.3% mAP50 and 28.2% mAP50–95, compared with 75.7% mAP50 on the daytime WIDER FACE images.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.