Abstract: A new method for single hazy traffic image enhancement is proposed. The relationships between original image gradient field and ideal image gradient field are obtained by calculating gradient on both sides of atmospheric scattering physical model, then the recovery of the ideal image is turned into extremum problem of functional about gradient energy, finally the partial differential equation model is obtained through variational?method. Using the dark channel prior, two transmission maps based patch dark channel and point dark channel can be estimated respectively, then each of them are processed by fast wavelet transform. Afterwards, the high-frequency component which comes from the transmission map of point dark channel and low-frequency component which comes from the transmission map of patch dark channel are extracted. Finally, a high quality ideal image can be obtained by solving the Euler equation by finite difference method. A comparative experiment with a few other state of the art algorithms shows that the algorithms can improve the quality of hazy image effectively, especially for hazy image which slants dark.

Key words: hazy traffic image, image enhancement, image degradation model, dark channel prior, partial differential equation

摘要： 提出了一种新的雾天交通图像增强算法，对大气物理散射模型两边同时取梯度，得到原始图像与无雾图像梯度场之间的关系，将无雾图像的恢复转化成梯度能量泛函求极值问题，由变分得到图像增强的偏微分方程模型。通过暗原色先验，得到块透射率和点透射率，用快速小波变换分解的方法将块透射率的低频和点透射率的高频融合计算得到透射率，最后用有限差分法求解欧拉方程获得无雾图像。仿真实验结果表明该算法能够有效提高图像的质量。

关键词: 雾天交通图像, 图像增强, 大气物理散射模型, 暗原色先验, 偏微分方程