结合暗通道先验与Hessian 正则项的图像去雾

在雾天拍摄户外图像，其对比度和可见度均受到严重的影响。目前图像去雾方法 通常依赖于准确的透射率图，而二阶的Hessian 正则项具有保留精细结构同时抑制阶梯伪影的 能力，可提高图像的对比度和可见度。为此采用暗通道先验方法获得有雾图像大气光值初始透 射率图，提出一种结合Hessian 正则项的二阶变分模型来细化初始透射率图及去雾图像。利用 交替方向乘子法(ADMM)，通过引入辅助变量，使拉格朗日乘子不断更新迭代，直到能量方程 收敛，输出去雾图像。采用LIVE Image Defogging 有雾图像数据库进行了仿真实验。通过对去 除薄雾和浓雾效果图的视觉质量和定量的评估，表明该方法得到的去雾图像清晰自然，纹理细 节保持效果较好。

Image dehazing combining dark channel prior and Hessian regular term

The contrast and visibility of outdoor images taken in hazy weather are seriously affected.At present,the image dehazing methods usually consider that the dehazing performance highly depend ends on the accurate transmission image.The second order Hessian regular term has the ability to preserve fine structure and suppress step artifacts,which is helpful to improve the image contrast and visibility.Therefore,in this paper,the dark channel prior method is first used to obtain atmospheric optical value and the initial transmission image,and then a second order variational model is proposed to refine the initial transmission image and dehazing image by combining Hessian regular term.In order to improve the operational efficiency of the proposed dehazing model,a corresponding alternating direction multiplier method(ADMM)was designed.By introducing auxiliary variables,the Lagrangian multiplier was continuously updated and iterated until the energy equation converged.At last,the simulation experiment was carried out by the foggy image data base(LIVE Image Defogging)to test the proposed fog removal method.The visual quality and quantitative evaluation of the effect pictures of mist and fog removal showed that the fog removal images obtained by the fog removal model proposed in this paper were clear and natural,and the texture details maintained well.