中文摘要：

为解决传统基于暗通道先验（DCP）的单幅去雾算法在景物边界处出现的白边现象及高计算复杂度,在中值DCP（MDCP）法的基础上提出一种改进算法。首先通过判断RGB三颜色通道的最小值与中值暗原色的差值绝对值的大小判断DCP法或MDCP法是否成立：当DCP法成立时,取较大窗口的暗原色值作为最终的暗原色值;当DCP法不成立而MDCP法成立时,取两种窗口中中值暗原色值较小的值作为最终的暗原色值;当DCP法和MDCP法都不成立时,取较小窗口的中值暗原色值作为最终的暗原色值。然后在HSI色彩空间保留色调分量不变,仅对强度分量运用修改的暗原色值进行去雾。最后,对饱和度分量进行颜色补偿。本文算法无需图像后期修补。实验结果表明了本文改进算法的可行性和有效性。

英文摘要：

The traditional single image dehazing algorithm based on dark channel prior （DCP） is accompa- nied with shortages such as the white border in the scene and the high computational complexity. In order to solve these problems, this paper proposes an improved algorithm based on the median dark channel prior （MI3CP）. Firstly,the absolute value of the minimum of RGB three color channels and the median dark channel are used to determine whether dark channel prior or median dark channel prior is applicable. If dark channel prior is applicable, the dark channel prior value of bigger window is decided as the final value of dark colors. Otherwise, the smaller median dark channel prior value of two windows is used to present the final value of dark colors. If neigher dark channe prior nor median dark channel prior is not applicable,the median dark channel prior value of smaller window is chosen as the final value of dark colors. Then it retains the hue in HSI color space, the intensity is defogged using the modified dark chan- nel values,and the saturation is improved to adopt color restoration. This method does not need the image post repairing. Experimental results show that the improved algorithm is effective and feasible.