Understanding Perceptual Quality in CCTV Images: A Benchmark Dataset and Entropy-based Insights

Yujin Han, Jiwoo Kang, Sanghoon Lee, Taewan Kim; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2025, pp. 3412-3419

Despite the increasing role of intelligent video surveillance in real-world analytics, existing image quality assessment (IQA) datasets rarely include authentic CCTV images, relying instead on artificially distorted or overly simplified natural scenes. Unlike general images, humans often perceive CCTV content as low quality even without synthetic distortions, due to factors such as poor luminance and sparse texture. This highlights the need for IQA datasets that reflect perceptual variability under operational CCTV conditions. To address this gap, we introduce CQA-DB, a dataset of 222 Full HD CCTV images collected from diverse indoor/outdoor and day/night environments. Subjective quality ratings from 53 participants yield a broad and balanced distribution of mean opinion scores. Notably, perceptual quality shows strong correlation with two proposed visual entropy metrics; one based on luminance and the other on texture complexity. Proposed dataset provides a high-fidelity benchmark for no-reference IQA in surveillance scenarios, and the dataset will be publicly released soon to support further research.