基于暗通道先验的快速图像去雾  

Fast Image Haze Removal Based on Dark Channel Prior

作　　者：王雪梅[1] 鞠铭烨[2] 

机构地区：[1]南京邮电大学计算机学院,南京210003 [2]南京邮电大学物联网学院,南京210003

出　　处：《科学技术与工程》2016年第20期66-72,共7页Science Technology and Engineering

基　　金：国家自然科学基金(61571241);江苏省研究生创新计划(CXZZ13_0476);南京邮电大学校级科研基金(NY215169)资助

摘　　要：为提高雾天降质图像的清晰度,基于暗通道先验提出单幅图像去雾的快速算法;其主要思想是摒弃复杂的软抠像方法,通过引入多分辨率处理,并结合最小滤波和联合双边滤波,对利用暗通道先验估计得到的透射图进行细化处理。针对天空区域暗通道先验失效引起色偏的问题,结合天空的特征先验和概率分布特征,识别出图像中的天空部分;并依据天空亮度来自适应调整透射率下限值。仿真表明:在保持良好去雾能力的同时,极大地降低了计算复杂度,满足一般工程的实时性要求。To enhance the visibility of degraded images acquired under poor weather conditions like haze and fog,a fast single image haze removal algorithm is proposed based on dark channel prior. The main idea of the algorithm is to refine the rough estimate of the transmission map via minimum filtering and joint bilateral filtering on basis of multi-resolution processing,rather than soft matting operation. In addition,to solve the problem of color shift in the sky region due to the invalidity of the dark channel prior,a novel adaptive method to determine the lower bound of transmission is designed,in which both the inherent features and the probability distribution of sky are considered to identify the sky region in the input image,and the transmission's lower bound is adjusted based on the intensity of sky region. The experimental results demonstrate that the proposed algorithm is capable of decrease the computational complexity significantly and meet the real-time requirement of realistic engineering tasks while achieve satisfactory dehazing effects as well.

关 键 词：暗通道先验 透射图 多分辨率 联合双边滤波 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]