单幅图像去雾的生成式方法研究综述

打开文本图片集

关键词：图像去雾；图像复原；生成式算法；扩散模型；生成对抗网络

中图分类号：TP391.41 文献标志码：A 文章编号：1001-3695（2026）01-003-0023-12

doi：10.19734/j. issn.1001-3695.2025.06.0198

Research review on generative methods for single image dehazing

Cao Qianwena†，Li Gongrub （a.ColegeofeyOeell，a Utou（n）， China）

Abstract：Image dehazingalgorithmsanalyzeand process hazyimages toremove foginterferenceandrestore clearandrealistic scene information，which iscrucial forenhancingtheperceptualrobustnessofcomputervisionsystemsundercomplex weatherconditions.Inrecentyears，theintroductionof generativemodelingconcepts intoimagedhazingtaskshas greatlyenrichedalgorithmicmodelsandlearningparadigms，graduallybecomingaresearchhotspotinthisfield.Tosystematicalyreview the current progress anddevelopment trendsof generative methods，this paper focused on single image dehazing and provided adetailedintroductionandanalysisofrepresentativegenerativedehazingapproaches，includingthosebasedonGAN，difusion model，and Transformer-fusion architecture.Subsequently，thepapersummarizedand compared mainstream image dehazing datasetsandcommonimagequalityevaluationmetrics.Fially，itdiscussdtheadvantagesandlimitationsof existingmethods.The studyalso explored future directions andchallenges of generative image dehazing.This paper aimed to provide theoretical support and structured references for related research in the field.

Keywords：imagedehazing；image restoration；generativealgorithm；difusionmodel；generativeadversarial network（GAN）

0 引言

根据人类对图像的主观视觉效果，恶劣天气可分为动态天气（如雨、雪、冰雹）和静态天气（如雾、霾）。（剩余33848字）