一种改进实时图像去雾的暗原色先验算法

A dark channel prior algorithm for improving real-time image dehazing

针对暗原色先验图像去雾算法运行时间长,天空区域分割耗时,复原图像中存在方块效应以及整体偏暗等缺点,提出一种改进的实时图像去雾暗原色先验算法。在暗通道求取时,使用快速最小值滤波,加快传统暗通道求取最小值排序的时间;在大气光求取时,使用像素值限定的方法快速排除天空区域,使用暗通道图像剩余最亮部分对应原图像像素值的均值,有效避免原算法中求取大气光值过大导致复原图像失真的现象;在透射率的优化时,使用Sobel边缘检测与求取图像二阶导数图像,得到精细的透射率,改善了细化透射率的时间复杂度;对最终图像采取伽码变换,提高图像亮度。实验结果表明,与原暗通道先验算法相比,此算法有效提高了算法的实时性。

In allusion to the disadvantages of the dark channel prior image defogging algorithm,such as long running time,time-consumption of sky region segmentation,and blocking effect and whole partial dark in the restored image,an improved real-time image defogging dark channel prior algorithm is proposed.In the dark channel calculation,the fast minimum filtering is used to speed up the minimum sorting of the traditional dark channel.In the atmospheric light extraction,the method of pixel value limit is used to quickly exclude the sky area,and make the remaining brightest part of the dark channel image correspond to the mean value of the original image pixel value,which can effectively avoid the restored image distortion caused by the too large atmospheric light value got by the original algorithm.In the optimization of transmissivity,the Sobel edge detection and second derivative image are used to obtain the refined transmissivity,which improves the time complexity of refining transmissivity.The Gamma conversion is applied to the final image to improve the brightness of the image.The experimental results show that in comparison with the original dark channel prior algorithm,this algorithm can effectively improve its realtime performance.