光学精密工程2025，Vol.33Issue(13)：2124-2135,12.DOI:10.37188/OPE.20253313.2124
面向多模式图像的改进暗通道先验去雾增强
Improved dark channel prior dehazing enhancement for multi-modal images
摘要
Abstract
A multi-modal image dehazing enhancement algorithm based on the dark channel prior is pro-posed to address the limitations of single-mode processing and restricted generality in existing image en-hancement methods.This algorithm is applicable to various polarization image modalities,including polar-ization intensity,Stokes parameters,and linear polarization images,as well as conventional RGB and grayscale images.For polarization images,atmospheric light estimation is performed through K-means clustering,grid partitioning,and bilinear interpolation,while atmospheric transmission is derived using brightness and structure-weighted techniques.Dark channel computation incorporates multi-scale Gauss-ian filtering combined with gradient-based adaptive weight fusion.For RGB and grayscale images,atmo-spheric light is estimated by K-means clustering and the 95th percentile of sky pixels,and atmospheric transmission is calculated via Gaussian Laplacian edge detection and bilinear interpolation.Dark channel computation utilizes multi-scale erosion operations alongside local contrast-based weighting.Experimental evaluation was conducted using multi-modal images collected under outdoor light mist and indoor artificial thick fog conditions,with dehazing enhancement outcomes compared against conventional dark channel prior and multi-scale Retinex algorithms.The results reveal marked improvements in image clarity,edge definition,and detail restoration.Specifically,polarization images demonstrated minimum enhancements of 112.6%,14.0%,and 5.0%in average gradient,image entropy,and peak signal-to-noise ratio,re-spectively,relative to the multi-scale Retinex algorithm.Non-polarization images exhibited minimum im-provements of 103.6%,20.6%,and 21.9%across the same metrics.This comprehensive validation confirms that the proposed algorithm not only significantly enhances image quality but also maintains ro-bust generality across diverse image modalities.关键词
图像去雾/图像增强/偏振/暗通道先验/多尺度Retinex算法Key words
image defogging/image enhancement/polarization/dark channel prior/multi-scale Retinex algorithm分类
信息技术与安全科学引用本文复制引用
卜祥涛,宋亚芳,王晓宇,姜珊,李德胜,赵宇,李亚红..面向多模式图像的改进暗通道先验去雾增强[J].光学精密工程,2025,33(13):2124-2135,12.基金项目
国家自然科学基金资助项目(No.11904044) （No.11904044）
辽宁省教育厅面上项目(No.JYTMS20230424) （No.JYTMS20230424）
