一种改进的基于暗原色理论的去雾方法  

An improved dehazing algorithm based on dark channel prior

作　　者：冯骢[1] 达飞鹏[1] 陈璋雯[1] 

机构地区：[1]东南大学自动化学院,南京210096

出　　处：《东南大学学报（自然科学版）》2012年第A01期70-73,共4页Journal of Southeast University：Natural Science Edition

基　　金：国家自然科学基金资助项目(51175081;61107001);江苏省自然科学基金资助项目(BK2010058)

摘　　要：针对有雾图像的浅色或偏白色区域无法正确估计透射率,从而导致图像的去雾及恢复产生过度增强的问题,提出了一种改进的基于暗原色先验的去雾方法.该方法首先提取图片中的浅色区域,并对其邻域场景分区,通过对比度分析来判断每个邻域区域与浅色区域场景的透射率的相似程度.相似度较高的区域在估计中的权重会得到增加.最后用暗原色算法获取临域区域的透射率信息,根据权重分配从而正确估计浅色区域的透射率.该方法可有效估计浅色区域的透射率,与现有方法相比,克服了其对浅色区域恢复失真的问题,去雾后的场景更为自然.实验结果表明,该方法切实可行且具有较好的鲁棒性,可达到良好的去雾效果.For the problem that the transmission of light color area in hazed image cannot be accurately obtained which may cause a overly enhanced restoration, a novel improved dehazing method based on dark channel prior is proposed. The light color area is extracted firstly and its adjacent area is divided into several regions. Contrast analysis is applied to judge the similarity between the light color area and each adjacent region. The region with high similarity is assigned with high weight in the estimation. Moreover,dark channel prior is used to obtain the transmission in each adjacent region. At last, the transmission of the light color area is estimated according to the weight assignment with the help of the transmissions of its adjacent regions. Compared with the current methods, the proposed method can re move the distortion and the recovered image is more natural. Experiments show that this method is ef fective and has better robustness which can obtain the accurate restoration.

关 键 词：去雾算法 暗原色先验 透射率 对比度 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]