Structure Interaction Dehazing Network Combined with YCbCr Color Space for Real-World Image Dehazing

Fengnian Zhao, Kai Lv
Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, PMLR 278:132-143, 2025.

Abstract

Dehazing in the RGB space often causes artifacts and detail blurring due to the difficulty in separating luminance and color. Traditional encoder-decoder models also suffer from semantic discontinuity and insufficient multi-scale feature interaction. To address these issues, we propose a novel method called the Structure Interaction Dehazing Network (SIDN), which leverages the advantages of the YCbCr color space in separating color and luminance to guide RGB feature extraction. SIDN consists of two core components: the Dual-space Feature Branch (DFB) and the Cross-Feature Block (CFB). The DFB integrates YCbCr features through the Phase Fusion Module (PFM) and Density-aware Feature Extraction Block (DFEB), enhancing texture recovery and guiding RGB feature reconstruction. The CFB improves multi-scale feature interaction and semantic alignment through an enhanced cross-layer mechanism. Experimental results show that SIDN achieves a 1.25dB improvement in PSNR and SSIM metrics on real-world datasets compared to previous methods, and also outperforms the latest methods in terms of FADE and NIQE metrics.

Cite this Paper

BibTeX
@InProceedings{pmlr-v278-zhao25a, title = {Structure Interaction Dehazing Network Combined with YCbCr Color Space for Real-World Image Dehazing}, author = {Zhao, Fengnian and Lv, Kai}, booktitle = {Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing}, pages = {132--143}, year = {2025}, editor = {Zeng, Nianyin and Pachori, Ram Bilas and Wang, Dongshu}, volume = {278}, series = {Proceedings of Machine Learning Research}, month = {25--27 Apr}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v278/main/assets/zhao25a/zhao25a.pdf}, url = {https://proceedings.mlr.press/v278/zhao25a.html}, abstract = {Dehazing in the RGB space often causes artifacts and detail blurring due to the difficulty in separating luminance and color. Traditional encoder-decoder models also suffer from semantic discontinuity and insufficient multi-scale feature interaction. To address these issues, we propose a novel method called the Structure Interaction Dehazing Network (SIDN), which leverages the advantages of the YCbCr color space in separating color and luminance to guide RGB feature extraction. SIDN consists of two core components: the Dual-space Feature Branch (DFB) and the Cross-Feature Block (CFB). The DFB integrates YCbCr features through the Phase Fusion Module (PFM) and Density-aware Feature Extraction Block (DFEB), enhancing texture recovery and guiding RGB feature reconstruction. The CFB improves multi-scale feature interaction and semantic alignment through an enhanced cross-layer mechanism. Experimental results show that SIDN achieves a 1.25dB improvement in PSNR and SSIM metrics on real-world datasets compared to previous methods, and also outperforms the latest methods in terms of FADE and NIQE metrics.} }
Endnote
%0 Conference Paper %T Structure Interaction Dehazing Network Combined with YCbCr Color Space for Real-World Image Dehazing %A Fengnian Zhao %A Kai Lv %B Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing %C Proceedings of Machine Learning Research %D 2025 %E Nianyin Zeng %E Ram Bilas Pachori %E Dongshu Wang %F pmlr-v278-zhao25a %I PMLR %P 132--143 %U https://proceedings.mlr.press/v278/zhao25a.html %V 278 %X Dehazing in the RGB space often causes artifacts and detail blurring due to the difficulty in separating luminance and color. Traditional encoder-decoder models also suffer from semantic discontinuity and insufficient multi-scale feature interaction. To address these issues, we propose a novel method called the Structure Interaction Dehazing Network (SIDN), which leverages the advantages of the YCbCr color space in separating color and luminance to guide RGB feature extraction. SIDN consists of two core components: the Dual-space Feature Branch (DFB) and the Cross-Feature Block (CFB). The DFB integrates YCbCr features through the Phase Fusion Module (PFM) and Density-aware Feature Extraction Block (DFEB), enhancing texture recovery and guiding RGB feature reconstruction. The CFB improves multi-scale feature interaction and semantic alignment through an enhanced cross-layer mechanism. Experimental results show that SIDN achieves a 1.25dB improvement in PSNR and SSIM metrics on real-world datasets compared to previous methods, and also outperforms the latest methods in terms of FADE and NIQE metrics.
APA
Zhao, F. & Lv, K.. (2025). Structure Interaction Dehazing Network Combined with YCbCr Color Space for Real-World Image Dehazing. Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, in Proceedings of Machine Learning Research 278:132-143 Available from https://proceedings.mlr.press/v278/zhao25a.html.

Related Material