文献类型：期刊文献

中文题名：融合雾线先验与对比度增强的图像去雾研究

英文题名：Research on Image Dehazing by Integrating Haze Line Prior and Contrast Enhancement

作者：景竑元[1];陈嘉星[1];周昊[1];宏晨[1];陈艾东[1]

第一作者：景竑元

机构：[1]北京联合大学机器人学院,北京100020

第一机构：北京联合大学机器人学院

年份：2024

卷号：38

期号：4

起止页码：51-60

中文期刊名：北京联合大学学报

外文期刊名：Journal of Beijing Union University

基金：北京市教育委员会科研计划项目(KM202111417008);北京联合大学科研项目(ZK20202401、ZK80202005);北京联合大学教改项目(JJ2021Q006)。

语种：中文

中文关键词：单幅图像去雾;雾线先验;软抠图;非均质雾;Retinex理论

外文关键词：single image dehazing;fog line prior;soft matting;nonhomogeneous haze;Retinex theory

摘要：经传统大气散射模型及雾线先验算法处理后的图像出现浓雾区域边缘模糊、物体边缘有光晕及颜色失真等现象,由此提出一种融合雾线先验与对比度增强技术的去雾算法。该算法首先基于雾线先验模型粗略估计有雾图像的透射率系数;然后采用软抠图技术对粗略估计的结果进行校准,得到相对准确的透射率,并选取区域的亮点作为大气环境光值,通过大气散射模型获得初步图像;再结合边缘信息,通过高斯滤波器对初步结果进行优化;最后用Frankle-McCann Retinex算法调节图像的对比度,获得优化后的去雾图像。实验结果表明:在客观评价结果中,与雾线先验算法相比,PSNR值平均提升了1.63左右,SSIM值平均提升了3%左右。由此可知,该算法与传统去雾算法相比,图像去雾效果显著,在改善图像细节的同时,也减小了图像色彩失真的影响。
After processing with traditional atmospheric scattering models and haze line prior algorithms,images often exhibit phenomena such as blurred edges in dense fog areas,halo effects around object edges,and color distortion.Consequently,a dehazing algorithm that integrates haze line prior with contrast enhancement techniques has been introduced.This algorithm initially provides a rough estimation of the transmission projection ratio coefficient of the hazy image based on the haze line prior model.Subsequently,soft matting technology is applied to calibrate the preliminary estimation results,yielding a relatively accurate transmittance.The bright spots within the region are chosen as the atmospheric light value,and a preliminary image is obtained through the atmospheric scattering model.Then,by incorporating edge information and employing a Gaussian filter,the preliminary results are optimized.Finally,the Frankle-McCann Retinex algorithm is used to adjust the contrast of the image,resulting in an optimized dehazed image.Experimental results show that,in objective evaluation metrics,the PSNR values have increased by an average of about 1.63,and the SSIM values have improved by an average of about 3%,compared to the haze line prior algorithms.This indicates that compared to existing mainstream traditional dehazing algorithms,this algorithm significantly improves the dehazing effect of images,enhancing image details while reducing the impact of color distortion.