Abstract
Low-Light Image Enhancement (LLIE) aims to recover visually pleasing content and details from degraded low-light images. However, existing RGB-based methods often suffer from color bias and brightness artifacts due to inherent high color sensitivity. Although the HSV color space can decouple brightness and color, it introduces noticeable red and black noise artifacts. To address these challenges, we adopt the Horizontal/Vertical-Intensity (HVI) color space for LLIE, which is defined by the HV color map and learnable intensity. The former enforces small distances for red coordinates to alleviate red noise artifacts, while the latter adaptively compresses low-light regions to suppress black noise artifacts. Based on the HVI space, we further propose HVI-CIDNet+, a Color and Intensity Decoupling Network designed to restore degraded content and mitigate color distortion in extremely dark regions. Specifically, HVI-CIDNet+ extracts abundant contextual and degradation-aware priors from low-light images using pre-trained vision-language models and integrates them through a novel Prior-guided Attention Block (PAB). Within PAB, latent semantic priors facilitate content restoration, while degraded representations guide more reliable color correction through a carefully designed cross-attention fusion mechanism. Furthermore, we introduce a Region Refinement Block that combines convolution for information-rich regions with attention for information-scarce regions, enabling accurate brightness adjustment. Extensive experiments on 10 benchmark datasets demonstrate that HVI-CIDNet+ consistently outperforms existing state-of-the-art methods. The code is available at the https://github.com/shikangbiao/CIDNet_extension
| Original language | English |
|---|---|
| Pages (from-to) | 12126-12139 |
| Number of pages | 14 |
| Journal | IEEE Transactions on Circuits and Systems for Video Technology |
| Volume | 36 |
| Issue number | 8 |
| DOIs | |
| State | Published - 1 Aug 2026 |
Keywords
- HVI color space
- degraded representations
- latent semantic priors
- low-light image enhancement
- region refinement
Fingerprint
Dive into the research topics of 'HVI-CIDNet+: Beyond Extreme Darkness for Low-Light Image Enhancement'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver