基于暗通道先验和核回归的图像去雾研究  

Research of image dehazing based on darkness-channel prior knowledge and kernel regression

作　　者：乔伟伟[1,2] 谢从华[2] 刘永俊[2] 王晓楠[2] 姚宇峰[2] 

机构地区：[1]苏州大学计算机科学与技术学院,江苏苏州215006 [2]常熟理工学院计算机科学与工程学院,江苏苏州215500

出　　处：《计算机应用研究》2017年第4期1277-1280,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(61402204);江苏省自然科学基金资助项目(BK2012209;BK20130529);苏州市科技发展计划项目(SYG201409)

摘　　要：针对雾霾天气条件下,大气粒子的散射作用导致的图像质量下降问题,提出一种基于暗通道先验知识与局部多项式核回归算法相结合的去雾方法。根据暗通道先验原理估计出大气光强度和初始透射率,采用局部多项式核回归对透射率进行精细化处理,利用细化后的透射率和估计出的大气光强度恢复雾霾图像。通过大量的实验数据表明,该方法可以有效地对雾霾图像实现去雾处理。与最新的几种去雾算法相比较,该方法处理后的图像保留了更多的细节信息,极大地提高了图像的清晰度。In view of the problem that atmospheric scattering degraded image quality in the hazy weather, this paper proposed a research of image dehazing based on darkness-channel prior knowledge and kernel regression theory. Firstly, it estimated atmospheric light intensity and initial transmission rate in terms of darkness channel prior theory. Secondly, it adopted the kernel regression method to refine initial transmission rate. Finally, it used transmission rate refined and atmospheric light intensity es- timated to restore the haze image. According to a large number of experimental data,it shows that this research can remove the fog of image effectively. Compared with the most advanced methods, the image processed by this algorithm not only retains more details of the original image, but also improves the image definition largely.

关 键 词：暗通道先验 核回归 透射率 大气光强度 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]