基于SLIC和SVR的单幅图像去雾算法  

Single image dehazing algorithm based on SLIC and SVR

作　　者：栾中[1,2] 尚媛园[1,2] 舒华忠[3] 周修庄 丁辉[1,2] 

机构地区：[1]首都师范大学信息工程学院,北京100048 [2]首都师范大学成像技术北京市高精尖创新中心,北京100048 [3]东南大学影像科学与技术实验室,南京210096

出　　处：《东南大学学报（自然科学版）》2018年第1期25-29,共5页Journal of Southeast University：Natural Science Edition

基　　金：国家自然科学基金资助项目(NSF61601311);北京市优秀人才培养资助项目(2016000020124G088);北京市长城学者资助项目(CIT&TCD20170322);北京市建设创新型团队和高校教师职业发展资助项目(IDHT20150507);首都师范大学青年科研创新团队资助项目

摘　　要：为了提升图像去雾算法的效果和适应性,提出了一种基于简单线性迭代聚类(SLIC)超像素分割和支持向量回归(SVR)的单幅图像去雾算法.首先,将雾霾图像进行超像素分割,对于每个超像素块,取其最大通道内的最大值作为局部大气光的估计,并使用引导滤波优化大气光图的边缘.然后,采集清晰户外图像并切割成块,利用大气散射模型为其加入不同程度的雾作为训练样本,提取对比度、饱和度、直方图均衡度和最小通道均值4种特征,利用支持向量回归算法训练传输参数估算模型.实验结果表明,所提算法有效地恢复了图像的对比度和颜色饱和度,同时无论是在主观视觉效果方面,还是在结构相似度和峰值信噪比等客观评价指标方面,所提算法均优于现有传统算法.To improve the effect and the adaptability of the dehazing algorithm,a single image dehazing algorithm based on simple linear iterative clustering（ SLIC） super-pixel segmentation and support vector regression（ SVR） is proposed. First,the haze images are segmented by super-pixel. As for each super pixel block,the local atmospheric light is estimated by the maximum value of the block maximum channel. The edge of the atmospheric light map is optimized by the guided filter.Then,the clear outdoor images are collected and split into blocks. By using the atmospheric scattering model,different levels of haze are added to these blocks to obtain the training samples. Four features including the contrast,the saturation,the histogram equalization,and the mean value of minimum channel are extracted. The transmission parameter estimation model is trained by SVR. The experimental results show that the proposed algorithm achieves good restoration for contrast and saturation. This algorithm outperforms the existing algorithms in the subjective visual effect and the objective evaluation indices such as the structural similarity and the peak signal-to-noise ratio.

关 键 词：图像恢复 图像去雾 动态大气光 支持向量机 超像素分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]