结合图像分割的暗原色先验去雾算法  

Image Dehazing Algorithm Combining Dark Channel Prior with Segmentation

作　　者：许丽红[1] 王敬东[1] 邱玉娇[1] 王子瑞[1] 

机构地区：[1]南京航空航天大学自动化学院,南京210016

出　　处：《光电子技术》2013年第1期27-35,共9页Optoelectronic Technology

摘　　要：基于暗原色先验规律的图像去雾算法在进行暗原色通道计算时,用固定区域会造成介质透射率估算的不合理,从而导致最终去雾效果细节不清晰和在较远景深场景模糊的现象。采用基于图论的分割算法来确定取暗原色通道的区域,能根据景深的变化自适应地改变区域大小,避免了固定区域求取暗原色的带来的介质透射率估计不合理问题。通过实验分析和验证,证实了该算法能有效地改善雾天图像的退化现象和提高图像清晰度。The classical method based on dark channel prior only uses fixed patch size to cal- culate the dark channel. This operation makes the estimated transmission map inaccurate and the dehazed image details unsharp with fuzziness in deep scene depth. The graph-cut segmentation al-gorithm is adopted to confirm the patch size of dark channel and the area size can be changed ac-cording to the variation of depth of field adaptively. The inaccuracy of fixed patch size is thus re-duced with more accurate estimated transmission map gained. Experimental results show that haze image degradation and the image sharpness are all improved effectively with this algorithm.

关 键 词：图像去雾 暗原色先验规律 基于图论的分割 介质透射率 

分 类 号：TN911.73[电子电信—通信与信息系统]