基于散射模型的雾天图像快速复原算法

Fast restoration method for foggy images based on atmosphere scattering model

基于暗原色先验的图像去雾算法能够较好地复原雾天图像,复原的结果清晰自然,但原始算法存在计算复杂度高、大气光估计不够准确、天空区域易出现失真以及无法处理偏色雾霾等缺点。从以上4个问题出发,以大气散射模型为基础,提出了新的估计透射率与大气光的方法。利用线性对比度拉伸对原含雾图像进行预处理,以消除偏色雾霾的影响,利用设置反馈参数的均值滤波估算雾天图像的大气透射图,通过建立权重图划分天空区域并确定大气光,根据天空区域的面积自适应地修正该区域的透射率,将其代入复原模型得到去雾图像。实验证明,该算法能够较好地克服原始算法的缺点,在降低算法复杂度的同时起到良好的复原效果。

The algorithm of single image haze removal using dark channel prior has a good result on restoring foggy images. But the original algorithm has some disadvantages such as the estimation of the global atmospheric light being not accurate enough and the transmission of some regions whose values being similar to the atmospheric light are underestimated. Be- sides, it could not restore images which are degraded by dust. Based on the atmosphere scattering model, this paper uses the contrast-stretching in order to keep the fog pure-white, then uses the mean filter with feedback parameters to acquire the transmission map. Finally, we build a weight map to obtain the accurate atmospheric light value and correct the wrong transmission in sky or white regions. Experiments and comparisons show that this method can overcome the disadvantages of the original algorithm and generates better results with low computation complexity.