改进的基于暗原色先验的图像去雾算法  

An Improved Algorithm of Haze Removal Based on Dark Channel Prior

作　　者：李坤[1,2] 兰时勇[1,2] 张建伟[1,2] 李璐[1,2] 

机构地区：[1]四川大学计算机学院,四川成都610065 [2]四川大学视觉合成图形图像技术国防重点学科实验室,四川成都610065

出　　处：《计算机技术与发展》2015年第2期6-11,共6页Computer Technology and Development

基　　金：国家"863"高技术发展计划项目(2013AA013802)

摘　　要：针对暗原色先验估计透射率偏小,不适用于天空区域,致使复原图像色彩失真、亮度偏暗等问题,提出了一种基于天空区域分割的图像去雾算法。通过对天空和非天空区域的雾天图像进行有区分地复原,克服了因不满足先验规律而导致的失真问题,通过Gray World和White Patch假设对复原图像亮度进行校正处理,解决了图像偏暗的不足。此外,通过在天空区域估计大气光值,使求得的介质透射率更为精确,复原出的细节更加清晰。实验结果表明,该算法能有效解决天空区域失真问题,复原后的图像也具有较高的亮度和对比度。在时间复杂度不显著增加的条件下,能够达到良好的去雾效果。Dark channel prior often gives a smaller value in transmission estimation, and doesn＇ t be suitable for the sky region, which leads to a dim image with cross-color sky regions after haze removal. To solve these problems, a new algorithm of haze removal based on sky region segmentation is proposed, which separates the sky region firstly, and use different methods to recover the scene radiance in sky and non-sky regions,overcoming the cross-color caused by failure of the prior rule. With a global White Patch adaptation in sky region and Gray World adaptation in non-sky region,it yields a brighter result, which is a good solution to solve the deficiency of the restored image＇ s dim. In addition,with estimating the atmospheric light only in sky regions,the method gives a more precise transmission and resuits in more details in the final restored image. Experimental results show that this method can handle the cross-color of sky regions effectively, and the restored result keeps sufficient brightness and contrast. It yields a better result with no much cost in time complexity.

关 键 词：去雾 暗原色先验 引导滤波 天空分割 白平衡 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]