基于暗原色先验的雾天图像清晰度复原

Haze image restoration based on dark channel prior

由于去雾是一个病态问题,对还原清晰化图像带来了挑战。本文基于大气散射物理模型提出了一种有效而精准的方法。雾天图像中天空区域对图像暗通道的求取存在很大的干扰,采用阈值分割方法结合二叉树策略能够快速而精准的定位并估计大气光值,进而结合维纳滤波将其应用在透射率的优化过程中,能够缓解尖锐边缘部分周围的晕轮效应,再通过形态学处理方法进一步优化透射图的边缘。通过大量户外雾天图像的测试结果表明,改进算法效率高,去雾效果好。

Defogging is a pathological problem of seeking unknown parameters,and reducing and sharpening the image which often produces the halo effect. An effective and accurate method is proposed based on the physical model of atmospheric scattering. In the fog image,the sky area has a great interference to the dark channel of the image. In this paper,the threshold segmentation method combined with the binary tree strategy can locate and estimate the atmospheric light quickly and accurately,and apply it to the transmissivity during optimization,the halo effect around the sharp edge can be mitigated,and then the edge of the transmission graph can be further optimized by morphological processing. The test results of a large number of outdoor fog images show that the improved method in this paper can obtain better experimental results.