摘要
目前,物理模型的单幅图像去雾已成为图像去雾算法研究的重点。在分析了暗原色先验知识的单幅图像去雾算法基础上,针对暗原色先验去雾算法时间复杂度大的缺点,比较了目前已有的暗原色先验改进去雾算法,提出了一种新的暗原色先验单幅图像去雾改进算法。通过引入快速、各向同性的低通高斯滤波器,实现对透射率图的平滑均匀,以代替暗原色去雾方法中精妙但时间复杂度高的软抠图算法;对于图像中图层交界处,提出了以区域中值滤波方法进行修正的算法,以及满足自适应要求的全局大气光求解详细算法。实验结果表明,结合了以上3点改进的快速去雾算法在保证图像去雾效果的同时,能大幅度提高暗原色去雾算法的速度,适用于对工程上的图像、视频实时去雾。
Currently, defogging algorithms based on the physical model of a single image become the focus of defogging researches. Compare several classical single image defogging algorithms, the defogging algorithm based on the dark channel prior knowledge of a single image is the most effective and appropriate method. Since the dark channel prior defogging algorithm has high time complexity and space complexity, there are many researchers accordingly contributed significant improvements to reduce the complexity and improve its efficiency. Comparing these improved algorithms and studying the advantages and disadvantages of defogging, we proposed a new dark channel prior defogging fast algorithm for single image. First, through the introduction of the fast, efficient and low-pass Gaussian filter to substitute the soft matting algorithm or other wave filter, we achieved a smooth and refined transmittance figure. Next, during the process of defogging, since the dark colors in the image at the border of different depth of fields may appear a white border phenomenon, we proposed an area median filtering method to adjust its impact. Finally, the detailed algorithm adaptive to meet the requirements of a global atmospheric optical image were presented. Experimental results showed that the improved algorithm based on single image with the combination of the above mentioned three steps can quickly reduce the fog effect from the original image to ensure the quality of the image, while greatly improve the speed of dark channel prior defogging algorithms. The improved method is efficient in pratical, for example in engineering images defogging process and in video real-time defogging.
出处
《地球信息科学学报》
CSCD
北大核心
2015年第4期494-499,共6页
Journal of Geo-information Science
基金
国家自然科学基金项目(61272351)
福建省教育厅A类项目(JA14312)
关键词
去雾
暗原色先验
大气光
透射率
高斯滤波器
defogging
dark channel prior
atmospherics
transmittance
Gaussian filter
