Back

HVI: A New Color Space for Low-light Image Enhancement

Qingsen YanYixu FengCheng ZhangGuansong PangKangbiao ShiPeng WuWei DongJinqiu SunYanning Zhang

article2025CVPR234 citations

Proposes a dedicated Horizontal/Vertical-Intensity color space alongside a lightweight decoupling network to eliminate red and black noise artifacts commonly found in low-light image enhancement.

Listen

Capturing visual data in low-light environments often yields severe noise, poor brightness, and color distortion. While traditional computer vision methods operate in standard red-green-blue (sRGB) color spaces, they frequently suffer from color bias because luminance and chrominance are tightly coupled. Alternative approaches that decouple brightness by converting images to hue-saturation-value (HSV) color spaces resolve illumination problems but introduce severe red color discontinuities and dark region artifacts. The article introduces a novel color space termed Horizontal/Vertical-Intensity (HVI) alongside a dedicated dual-branch neural network, the Color and Intensity Decoupling Network (CIDNet), to deliver robust, artifact-free low-light image enhancement.

The research demonstrates how the proposed HVI representation overcomes HSV's limitations by applying polarization to the hue axis to eliminate red discontinuities and introducing an adaptive intensity collapse function to cluster and suppress noise in near-black regions. The paired CIDNet architecture independently processes chromatic information and scene brightness across dual branches, utilizing cross-attention mechanisms to exchange guidance between features. Evaluated across 10 benchmark datasets, CIDNet achieved top-tier visual and quantitative performance, achieving an optimal balance between quality and computational efficiency with only 1.88 million parameters and 7.57 GFLOPs of computational load. Furthermore, applying the HVI transformation as a plug-and-play pre-processing module to six external state-of-the-art models consistently improved their enhancement quality, increasing peak signal-to-noise ratios by up to 3.56 decibels.

These results indicate that adopting the HVI color space substantially reduces computational overhead and operational risk for real-world computer vision deployments in low-light scenarios, outperforming heavier diffusion-based models while running at a fraction of the processing time. Organizations building computer vision pipelines should consider integrating the HVI transform and CIDNet framework to improve low-light image quality efficiently. Future efforts should focus on deploying these lightweight models in live edge environments and conducting pilot tests across specialized sensor hardware.

Northwestern Polytechnical UniversitySingapore Management UniversityXi'an University of Architecture and Technologyimage enhancementcolor componentlow-light image enhancementHVI color spacePerceptual-inverse HVI
arXiv: 2502.20272Fediory/HVI-CIDNet

No sufficiently relevant recommendations were found.

Cover for HVI: A New Color Space for Low-light Image Enhancement

Abstract

Low-Light Image Enhancement (LLIE) is a crucial computer vision task that aims to restore detailed visual information from corrupted low-light images. Many existing LLIE methods are based on standard RGB (sRGB) space, which often produce color bias and brightness artifacts due to inherent high color sensitivity in sRGB. While converting the images using Hue, Saturation and Value (HSV) color space helps resolve the brightness issue, it introduces significant red and black noise artifacts. To address this issue, we propose a new color space for LLIE, namely Horizontal/Vertical-Intensity (HVI), defined by polarized HS maps and learnable intensity. The former enforces small distances for red coordinates to remove the red artifacts, while the latter compresses the low-light regions to remove the black artifacts. To fully leverage the chromatic and intensity information, a novel Color and Intensity Decoupling Network (CIDNet) is further introduced to learn accurate photometric mapping function under different lighting conditions in the HVI space. Comprehensive results from benchmark and ablation experiments show that the proposed HVI color space with CIDNet outperforms the state-of-the-art methods on 10 datasets. The code is available at https://github.com/Fediory/HVI-CIDNet.

Topics

image enhancementcolor component

Keywords

low-light image enhancementHVI color spacePerceptual-inverse HVIbrightness enhancementadaptive intensity collapsecolor distortionimage restorationsRGB color space

Table of Contents

  • 1. Introduction
  • 2. Related Work
  • 2.1. Low-Light Image Enhancement
  • 2.2. Color Space
  • 3. HVI Color Space
  • 3.1. Color Space Noises in HSV
  • 3.2. Horizontal/Vertical Plane with Polarized HS and Collapsible Intensity
  • 4. Color and Intensity Decoupling Network
  • 4.1. HVI Transformation
  • 4.2. Dual-branch Enhancement Network
  • 4.3. Perceptual-inverse HVI Transformation
  • 4.4. Loss Function
  • 5. Experiments
  • 5.1. Datasets and Settings
  • 5.2. Main Results
  • 5.3. Ablation Study
  • 6. Conclusion
  • Acknowledgement
  • References

Knowls

  1. Knowl 1 — Polarized hue–saturation coordinates remove the HSV red discontinuity

    model/method

    For each pixel, let H(x)∈[0,6)H(x)\in[0,6) and S(x)∈[0,1]S(x)\in[0,1] denote the hue and saturation obtained from the HSV representation of an sRGB image. HVI replaces the discontinuous hue coordinate with orthogonal polarized coordinates:

    PH(x)=cos⁡(πH(x)3),PV(x)=sin⁡(πH(x)3).P_H(x)=\cos\left(\frac{\pi H(x)}{3}\right),\qquad P_V(x)=\sin\left(\frac{\pi H(x)}{3}\right).

    The chromatic coordinates are S(x)PH(x)S(x)P_H(x) and S(x)PV(x)S(x)P_V(x). Because hue values H=0H=0 and H=6H=6 represent the same red color and are mapped to the same point (1,0)(1,0), nearby red colors remain nearby in the polarized plane instead of being separated at opposite ends of the HSV hue axis. This operation preserves HSV’s separation of chromaticity from intensity while reducing the red-discontinuity noise that produces artifacts during low-light enhancement.

  2. Knowl 2 — Learnable intensity collapse completes the HVI color space

    model/method

    Let the input sRGB image be I∈[0,1]H×W×3\mathbf I\in[0,1]^{H\times W\times3}, let xx be a pixel, and let Ic(x)I_c(x) be its channel value for c∈{R,G,B}c\in\{R,G,B\}. HVI uses the Max-RGB intensity map

    Imax⁡(x)=max⁡c∈{R,G,B}Ic(x).I_{\max}(x)=\max_{c\in\{R,G,B\}} I_c(x).

    For a trainable darkness-density parameter k∈Q+k\in\mathbb Q^+, the adaptive intensity-radius function is

    Ck(x)=[sin⁡(πImax⁡(x)2)+ε]1/k,ε=10−8.C_k(x)=\left[\sin\left(\frac{\pi I_{\max}(x)}{2}\right)+\varepsilon\right]^{1/k},\qquad \varepsilon=10^{-8}.

    If H(x)H(x) and S(x)S(x) are the HSV hue and saturation, HVI forms its horizontal and vertical chromatic channels as

    H^(x)=Ck(x)S(x)cos⁡(πH(x)3),V^(x)=Ck(x)S(x)sin⁡(πH(x)3).\widehat H(x)=C_k(x)S(x)\cos\left(\frac{\pi H(x)}{3}\right),\qquad \widehat V(x)=C_k(x)S(x)\sin\left(\frac{\pi H(x)}{3}\right).

    The HVI representation is the three-channel tuple (H^,V^,Imax⁡)(\widehat H,\widehat V,I_{\max}). The function CkC_k contracts the radius of dark colors toward zero while allowing brighter colors to approach radius one; learning kk adapts the amount of low-light-region collapse to the dataset and network. This suppresses HSV black-plane noise without discarding the intensity information needed for enhancement.

  3. Knowl 3 — Perceptual-inverse HVI transformation reconstructs adjustable HSV and sRGB images

    equation

    Given an enhanced HVI image (H^,V^,I^I)(\widehat H,\widehat V,\widehat I_I), CIDNet first removes the intensity-dependent radius using the same learned collapse function evaluated on I^I\widehat I_I:

    H~=H^Ck+ε,V~=V^Ck+ε,ε=10−8.\widetilde H=\frac{\widehat H}{C_k+\varepsilon},\qquad \widetilde V=\frac{\widehat V}{C_k+\varepsilon}, \qquad \varepsilon=10^{-8}.

    The perceptual-inverse HVI transformation then maps these intermediate coordinates to HSV values:

    HHSV=arctan⁡(V~H~) mod 1,H_{\mathrm{HSV}}=\arctan\left(\frac{\widetilde V}{\widetilde H}\right)\bmod 1, SHSV=αSH~2+V~2,VHSV=αII^I.S_{\mathrm{HSV}}=\alpha_S\sqrt{\widetilde H^2+\widetilde V^2},\qquad V_{\mathrm{HSV}}=\alpha_I\widehat I_I.

    Here αS\alpha_S and αI\alpha_I are user-configurable linear factors controlling saturation and brightness, respectively. Standard HSV-to-sRGB conversion produces the final enhanced image. The mapping is surjective and permits saturation and brightness to be adjusted independently after enhancement.

  4. Knowl 4 — CIDNet decouples HVI chromatic and intensity enhancement with cross-attention

    model/method

    The Color and Intensity Decoupling Network (CIDNet) processes an sRGB input in three stages: HVI transformation, dual-branch enhancement, and perceptual-inverse HVI transformation. Its enhancement module is a UNet-style encoder–decoder with skip connections and six Lighten Cross-Attention (LCA) modules, three in the encoder and three in the decoder.

    The intensity branch receives the HVI intensity channel and learns brightness enhancement over the whole image. The HV branch receives the concatenation of the HVI chromatic channels and the intensity channel, and learns chromatic denoising, particularly in dark regions where noise is strong. Cross-attention exchanges information between the branches rather than applying independent self-attention: intensity features guide chromatic denoising according to local illumination, while denoised intensity information helps smooth the brightness output. The enhanced HVI channels are converted back to sRGB by the perceptual-inverse HVI transformation. CIDNet has 1.88 million parameters and 7.57 GFLOPs for a 256×256256\times256 input.

  5. Knowl 5 — Joint HVI-space and sRGB-space supervision trains CIDNet

    equation

    Let I\mathbf I be the paired normal-light sRGB ground truth, IHVI\mathbf I_{HVI} its HVI transformation, I^\widehat{\mathbf I} the sRGB output of CIDNet, and I^HVI\widehat{\mathbf I}_{HVI} the enhanced HVI output before inverse transformation. CIDNet minimizes the combined loss

    L=λ ℓ(I^HVI,IHVI)+ℓ(I^,I),\mathcal L=\lambda\,\ell\left(\widehat{\mathbf I}_{HVI},\mathbf I_{HVI}\right)+\ell\left(\widehat{\mathbf I},\mathbf I\right),

    where ℓ(⋅,⋅)\ell(\cdot,\cdot) is the paper’s image discrepancy loss and λ\lambda weights HVI-space supervision relative to sRGB-space supervision. The HVI term constrains the learned output to the desired low-light color-space distribution, including the polarized red and collapsed dark regions, while the sRGB term preserves pixel-level structure and spatial detail.

  6. Knowl 6 — Training and evaluation protocol spans paired, extreme-dark, and unpaired LLIE benchmarks

    experimental setup

    CIDNet is evaluated on LOLv1, LOLv2-Real, LOLv2-Synthetic, SICE-Mix, SICE-Grad, Sony-Total-Dark, DICM, LIME, MEF, NPE, and VV. For LOLv1 and LOLv2-Real, training uses 400×400400\times400 crops, batch size 8, and 1,500 epochs. LOLv2-Synthetic uses batch size 1, 500 epochs, and no cropping. SICE uses 160×160160\times160 crops, batch size 10, and 1,000 epochs. Sony-Total-Dark uses 256×256256\times256 crops, batch size 4, and 1,000 epochs; its raw SID images are converted to sRGB without gamma correction to create an extreme-darkness setting.

    The model is implemented in PyTorch and trained with Adam using (β1,β2)=(0.9,0.999)(\beta_1,\beta_2)=(0.9,0.999) on one NVIDIA 2080Ti or 3090 GPU. The learning rate follows cosine annealing from 10−410^{-4} to 10−710^{-7}. Paired datasets are evaluated with PSNR, SSIM, and AlexNet-based LPIPS; unpaired datasets are evaluated with BRISQUE and NIQE.

  7. Knowl 7 — CIDNet achieves the strongest reported results on the LOL benchmarks with low complexity

    data/table

    On LOLv1, LOLv2-Real, and LOLv2-Synthetic, CIDNet is compared with representative high-performing baselines using PSNR in dB, SSIM, and LPIPS; higher PSNR and SSIM and lower LPIPS are better. FLOPs are measured for one 256×256256\times256 image.

    Could not parse LaTeX table

    CIDNet has the best value for every displayed metric on all three LOL variants. Relative to GSAD, it uses 10.8% as many parameters and far fewer FLOPs; relative to RetinexFormer, it reduces computation while obtaining higher scores on every displayed LOL benchmark.

  8. Knowl 8 — CIDNet improves performance on SICE and Sony-Total-Dark and obtains the best unpaired NIQE

    data/table

    The paired SICE and Sony-Total-Dark results use PSNR and SSIM, while the five unpaired datasets DICM, LIME, MEF, NPE, and VV are summarized with BRISQUE and NIQE. Higher PSNR and SSIM and lower BRISQUE and NIQE are better.

    Could not parse LaTeX table

    CIDNet is best on both paired metrics for SICE and Sony-Total-Dark, exceeding the second-best Sony-Total-Dark PSNR by 6.678 dB. It also achieves the best NIQE among the compared methods on the unpaired data, although RetinexNet has a slightly lower BRISQUE value than CIDNet.

  9. Knowl 9 — HVI functions as a plug-in color space for diverse LLIE architectures

    empirical result

    Replacing an sRGB input and output interface with the HVI transformation and its perceptual inverse improves several independent LLIE models on LOLv2-Real. The values in parentheses are the paper’s reported absolute changes relative to the corresponding sRGB implementation; their signs are reproduced exactly. GPU time is inference time in seconds.

    Could not parse LaTeX table

    The plug-in experiments show that HVI is not restricted to CIDNet: it improves PSNR, SSIM, or LPIPS across the tested sRGB-based methods, with the largest reported PSNR gain being 3.562 dB for GSAD. CIDNet has the shortest reported GPU time while also having the highest PSNR and the second-best SSIM and LPIPS among these entries.

  10. Knowl 10 — Ablations show that polarization, intensity collapse, dual branches, cross-attention, and dual-space loss are complementary

    data/table

    All ablations are conducted on LOLv2-Real using PSNR, SSIM, and LPIPS. For color-space comparisons, the paper uses a UNet with self-attention because the dual-branch and cross-attention design is specifically tailored to HVI.

    Could not parse LaTeX table

    Polarization alone removes red discontinuity artifacts, while CkC_k alone improves intensity handling but can confuse red with other colors and create artifacts outside red regions. Their combination gives the strongest color-space result. Adding a dual branch to self-attention raises PSNR by 0.846 dB over the single-branch self-attention variant, and adding cross-attention yields the full model’s best restoration. HVI-only supervision loses pixel-level structural consistency, whereas sRGB-only supervision neglects the desired HVI distribution; combining both losses is best on all three metrics.

Coverage note — No substantial contributed material was omitted; background, related work, acknowledgements, and references were excluded as non-contributory.

References

  1. 1.A. Brateanu, R. Balmez, A. Avram, and C. Orhei. Lyt-net: Lightweight yuv transformer-based network for low-light image enhancement, 2024. 3
  2. 2.Jianrui Cai, Shuhang Gu, and Lei Zhang. Learning a deep single image contrast enhancer from multi-exposure images. IEEE Transactions on Image Processing, 27(4):2049–2062, 2018. 5
  3. 3.Yuanhao Cai, Hao Bian, Jing Lin, Haoqian Wang, Radu Timofte, and Yulun Zhang. Retinexformer: One-stage retinex-based transformer for low-light image enhancement. In Proceedings of ICCV, pages 12504–12513, 2023. 1, 2, 6, 7
  4. 4.Chen Chen, Qifeng Chen, Jia Xu, and Vladlen Koltun. Learning to see in the dark. In CVPR, 2018. 5
  5. 5.Guo-Dong Fan, Bi Fan, Min Gan, Guang-Yong Chen, and C. L. Philip Chen. Multiscale low-light image enhancement network with illumination constraint. IEEE Transactions on Circuits and Systems for Video Technology, 32(11):7403–7417, 2022. 2
  6. 6.Yixu Feng, Shuo Hou, Haotian Lin, Yu Zhu, Peng Wu, Wei Dong, Jinqiu Sun, Qingsen Yan, and Yanning Zhang. Difflight: Integrating content and detail for low-light image enhancement. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6143–6152, 2024. 2, 7
  7. 7.James D Foley and Andries Van Dam. Fundamentals of interactive computer graphics. Addison-Wesley Longman Publishing Co., Inc., 1982. 3, 5
  8. 8.Zhenqi Fu, Yan Yang, Xiaotong Tu, Yue Huang, Xinghao Ding, and Kai-Kuang Ma. Learning a simple low-light image enhancer from paired low-light instances. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 22252–22261, 2023. 3, 4, 6
  9. 9.Theo Gevers, Arjan Gijsenij, Joost Van de Weijer, and Jan-Mark Geusebroek. Color in computer vision: fundamentals and applications. John Wiley & Sons, 2012. 1, 3, 7
  10. 10.Chunle Guo Guo, Chongyi Li, Jichang Guo, Chen Change Loy, Junhui Hou, Sam Kwong, and Runmin Cong. Zero-reference deep curve estimation for low-light image enhancement. In Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR), pages 1780–1789, 2020. 1, 6
  11. 11.Xiaojie Guo and Qiming Hu. Low-light image enhancement via breaking down the darkness. International Journal of Computer Vision, 131(1):48–66, 2023. 2, 3, 6
  12. 12.X. Guo, Y. Li, and H. Ling. Lime: Low-light image enhancement via illumination map estimation. IEEE Transactions on Image Processing, 26(2):982–993, 2017. 5, 6
  13. 13.Zhiquan He, Wu Ran, Shulin Liu, Kehua Li, Jiawen Lu, Changyong Xie, Yong Liu, and Hong Lu. Low-light image enhancement with multi-scale attention and frequency-domain optimization. IEEE Transactions on Circuits and Systems for Video Technology, 34(4):2861–2875, 2024. 2
  14. 14.Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In Advances in Neural Information Processing Systems, pages 6840–6851. Curran Associates, Inc., 2020. 2
  15. 15.Jinhui Hou, Zhiyu Zhu, Junhui Hou, Hui Liu, Huanqiang Zeng, and Hui Yuan. Global structure-aware diffusion process for low-light image enhancement. Advances in Neural Information Processing Systems, 36, 2024. 1, 6, 7
  16. 16.Jinhui Hou, Zhiyu Zhu, Junhui Hou, Hui Liu, Huanqiang Zeng, and Hui Yuan. Global structure-aware diffusion process for low-light image enhancement. Advances in Neural Information Processing Systems, 36, 2024. 2, 6
  17. 17.Hai Jiang, Ao Luo, Haoqiang Fan, Songchen Han, and Shuaicheng Liu. Low-light image enhancement with wavelet-based diffusion models. ACM Transactions on Graphics (TOG), 42(6):1–14, 2023. 2
  18. 18.Yifan Jiang, Xinyu Gong, Ding Liu, Yu Cheng, Chen Fang, Xiaohui Shen, Jianchao Yang, Pan Zhou, and Zhangyang Wang. Enlightengan: Deep light enhancement without paired supervision. IEEE Transactions on Image Processing, 30:2340–2349, 2021. 1, 2, 6
  19. 19.Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization, 2017. 5
  20. 20.Alex Krizhevsky, I. Sutskever, and G. Hinton. Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 25(2), 2012. 5
  21. 21.Edwin H Land. The retinex theory of color vision. Scientific american, 237(6):108–129, 1977. 3
  22. 22.Chulwoo Lee, Chul Lee, and Chang-Su Kim. Contrast enhancement based on layered difference representation of 2d histograms. IEEE Transactions on Image Processing, 22(12):5372–5384, 2013. 5, 6
  23. 23.Jaerin Lee, JoonKyu Park, Sungyong Baik, and Kyoung Mu Lee. Rethinking RGB color representation for image restoration models, 2024. 1
  24. 24.Chongyi Li, Chunle Guo, Linghao Han, Jun Jiang, Ming-Ming Cheng, Jinwei Gu, and Chen Change Loy. Low-light image and video enhancement using deep learning: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(12):9396–9416, 2022. 1, 4
  25. 25.Zhi Li, Zhenhong Jia, Jie Yang, and Nikola Kasabov. Low illumination video image enhancement. IEEE photonics journal, 12(4):1–13, 2020. 1
  26. 26.Kin Gwn Lore, Adedotun Akintayo, and Soumik Sarkar. Ll-net: A deep autoencoder approach to natural low-light image enhancement, 2016. 2
  27. 27.Ilya Loshchilov and Frank Hutter. Sgdr: Stochastic gradient descent with warm restarts, 2017. 5
  28. 28.Kede Ma, Kai Zeng, and Zhou Wang. Perceptual quality assessment for multi-exposure image fusion. IEEE Transactions on Image Processing, 24(11):3345–3356, 2015. 5, 6
  29. 29.Anish Mittal, Anush Krishna Moorthy, and Alan Conrad Bovik. No-reference image quality assessment in the spatial domain. IEEE Transactions on Image Processing, 21(12):4695–4708, 2012. 5
  30. 30.Anish Mittal, Rajiv Soundararajan, and Alan C. Bovik. Making a “completely blind” image quality analyzer. IEEE Signal Processing Letters, 20(3):209–212, 2013. 5
  31. 31.Olivier Petit, Nicolas Thome, Clement Rambour, and Luc ´ Soler. U-net transformer: Self and cross attention for medical image segmentation, 2021. 7, 8
  32. 32.Charles Poynton. 18 - image digitization and reconstruction. In Digital Video and HDTV, pages 187–194. Morgan Kaufmann, San Francisco, 2003. 3
  33. 33.Liu Risheng, Ma Long, Zhang Jiaao, Fan Xin, and Luo Zhongxuan. Retinex-inspired unrolling with cooperative prior architecture search for low-light image enhancement. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2021. 6
  34. 34.Steven A Shafer. Using color to separate reflection components. Color Research & Application, 10(4):210–218, 1985. 3
  35. 35.Chunyan She, Fujun Han, Lidan Wang, Shukai Duan, and Tingwen Huang. Mpc-net: Multi-prior collaborative network for low-light image enhancement. IEEE Transactions on Circuits and Systems for Video Technology, pages 1–1, 2024. 1
  36. 36.Vassilios Vonikakis, Rigas Kouskouridas, and Antonios Gasteratos. On the evaluation of illumination compensation algorithms. Multimedia Tools and Applications, 77:1–21, 2018. 5, 6
  37. 37.Chenxi Wang, Hongjun Wu, and Zhi Jin. Fourllie: Boosting low-light image enhancement by fourier frequency information. In Proceedings of the 31st ACM International Conference on Multimedia, pages 7459–7469, 2023. 7
  38. 38.Shuhang Wang, Jin Zheng, Hai-Miao Hu, and Bo Li. Naturalness preserved enhancement algorithm for non-uniform illumination images. IEEE transactions on image processing : a publication of the IEEE Signal Processing Society, 22, 2013. 5, 6
  39. 39.Tao Wang, Kaihao Zhang, Tianrun Shen, Wenhan Luo, Bjorn Stenger, and Tong Lu. Ultra-high-definition low-light image enhancement: A benchmark and transformer-based method. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 2654–2662, 2023. 6, 7
  40. 40.Wenjing Wang, Huan Yang, Jianlong Fu, and Jiaying Liu. Zero-reference low-light enhancement via physical quadruple priors. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 26057–26066, 2024. 2, 3, 6
  41. 41.Yufei Wang, Renjie Wan, Wenhan Yang, Haoliang Li, Lap-Pui Chau, and Alex C Kot. Low-light image enhancement with normalizing flow. arXiv preprint arXiv:2109.05923, 2021. 6
  42. 42.Zhou Wang, A.C. Bovik, H.R. Sheikh, and E.P. Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing, 13(4):600–612, 2004. 5
  43. 43.Chen Wei, Wenjing Wang, Wenhan Yang, and Jiaying Liu. Deep retinex decomposition for low-light enhancement. arXiv preprint arXiv:1808.04560, 2018. 2, 3, 4, 5, 6
  44. 44.Wenhui Wu, Jian Weng, Pingping Zhang, Xu Wang, Wenhan Yang, and Jianmin Jiang. Uretinex-net: Retinex-based deep unfolding network for low-light image enhancement. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5901–5910, 2022. 4, 6
  45. 45.Yuhui Wu, Guoqing Wang, Zhiwen Wang, Yang Yang, Tianyu Li, Peng Wang, Chongyi Li, and Heng Tao Shen. Recodiff: Explore retinex-based condition strategy in diffusion model for low-light image enhancement. arXiv preprint arXiv:2312.12826, 2023. 2
  46. 46.Xiaogang Xu, Ruixing Wang, Chi-Wing Fu, and Jiaya Jia. Snr-aware low-light image enhancement. In 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 17693–17703, 2022. 6, 7
  47. 47.Wenhan Yang, Wenjing Wang, Haofeng Huang, Shiqi Wang, and Jiaying Liu. Sparse gradient regularized deep retinex network for robust low-light image enhancement. IEEE Transactions on Image Processing, 30:2072–2086, 2021. 5
  48. 48.Xunpeng Yi, Han Xu, Hao Zhang, Linfeng Tang, and Jiayi Ma. Diff-retinex: Rethinking low-light image enhancement with a generative diffusion model. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 12302–12311, 2023. 2, 3
  49. 49.Syed Waqas Zamir, Aditya Arora, Salman Khan, Munawar Hayat, Fahad Shahbaz Khan, and Ming-Hsuan Yang. Restormer: Efficient transformer for high-resolution image restoration. In CVPR, 2022. 7, 8
  50. 50.Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In CVPR, 2018. 5
  51. 51.Yonghua Zhang, Jiawan Zhang, and Xiaojie Guo. Kindling the darkness: A practical low-light image enhancer. In Proceedings of the 27th ACM International Conference on Multimedia, pages 1632–1640, New York, NY, USA, 2019. ACM. 1, 2, 3, 6
  52. 52.Yu Zhang, Xiaoguang Di, Bin Zhang, Ruihang Ji, and Chunhui Wang. Better than reference in low-light image enhancement: Conditional re-enhancement network. IEEE Transactions on Image Processing, 31:759–772, 2021. 1
  53. 53.Shen Zheng, Yiling Ma, Jinqian Pan, Changjie Lu, and Gaurav Gupta. Low-light image and video enhancement: A comprehensive survey and beyond. arXiv preprint arXiv:2212.10772, 2022. 5
  54. 54.Dewei Zhou, Zongxin Yang, and Yi Yang. Pyramid diffusion models for low-light image enhancement. arXiv preprint arXiv:2305.10028, 2023. 2
  55. 55.Han Zhou, Wei Dong, Xiaohong Liu, Shuaicheng Liu, Xiongkuo Min, Guangtao Zhai, and Jun Chen. Glare: Low light image enhancement via generative latent feature based codebook retrieval. In Proceedings of the European Conference on Computer Vision (ECCV), 2024. 1
  56. 56.Libing Zhou, Xiaojing Chen, Baisong Ye, Xueli Jiang, Sheng Zou, Liang Ji, Zhengqian Yu, Jianjian Wei, Yexin Zhao, and Tianyu Wang. A low-light image enhancement method based on hsv space. The Imaging Science Journal, pages 1–14, 2023. 1
  57. 57.Shangchen Zhou, Chongyi Li, and Chen Change Loy. Lednet: Joint low-light enhancement and deblurring in the dark. In ECCV, 2022. 7

Citation

MLA
Yan, Q., et al. “HVI: A New Color Space for Low-light Image Enhancement”. arXiv, 2025, http://arxiv.org/abs/2502.20272v2.
APA
Yan, Q., Feng, Y., Zhang, C., Pang, G., Shi, K., Wu, P., Dong, W., Sun, J., & Zhang, Y. (2025). HVI: A New Color Space for Low-light Image Enhancement. arXiv. http://arxiv.org/abs/2502.20272v2
Chicago
Yan, Q., Y. Feng, C. Zhang, et al. 2025. “HVI: A New Color Space for Low-light Image Enhancement”. arXiv. http://arxiv.org/abs/2502.20272v2.
Harvard
Yan, Q. et al. (2025) “HVI: A New Color Space for Low-light Image Enhancement”, arXiv [Preprint]. Available at: http://arxiv.org/abs/2502.20272v2.
Vancouver
1. Yan Q, Feng Y, Zhang C, Pang G, Shi K, Wu P, Dong W, Sun J, Zhang Y (2025) HVI: A New Color Space for Low-light Image Enhancement. arXiv

BibTeX

@article{yan2025hvi,
  title = {HVI: A New Color Space for Low-light Image Enhancement},
  author = {Yan, Qingsen and Feng, Yixu and Zhang, Cheng and Pang, Guansong and Shi, Kangbiao and Wu, Peng and Dong, Wei and Sun, Jinqiu and Zhang, Yanning},
  year = {2025},
  journal = {arXiv},
  url = {http://arxiv.org/abs/2502.20272v2},
  eprint = {2502.20272}
}
Metadata:arXiv

Source Code

This paper has an official code repository available. Click below to access the source code.

View Repository

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF
License: IEEE