针对图像去雾领域缺乏有效雾浓度检测的不足,提出一种标准图像集广义灰度差-比的散点图先验,并引入Naka-Rushton函数拟合散点图实现图像雾浓度检测。首先,建立不同条件下标准图像集灰度散点图,提取广义灰度差-比先验；从图像像质退化模型出发,验证了灰度差-比先验的正确性；其次,建立符合先验约束的Naka-Rushton(NR)拟合函数, 计算标准图像集拟合NR函数的参数(n,k),建立(n,k)与视场像雾浓度对应的查找表；再次,采用回归分析的方法计算真实有雾图像散点图拟合参数(n",k")；计算标准图像集参数(n,k)与真实雾图像拟合参数(n",k")间的综合相关系数,搜索查找表并以过门限的综合相关系数所对应的(n,k)作为当前雾浓度评定等级。通过不同浓度有雾图像测试,证明算法测试结果符合浓度变化趋势；经过同场景不同浓度、不同场景不同浓度样本测试,算法测试结果与PM2.5相关系数达0.95,表明算法能够作为视场雾浓度等级评定；经过横向对比测试表明本文算法在测试精度达到4.8%。

Concerning the shortcoming of fog-concentration inspection in the field of image defogging, a scatterplot prior of generalized pixel difference-ratio (GPDR) was proposed and Naka-Rushton fitting function was introduced to inspect the fog concentration. Firstly, the gray scatterplots were built under standard foggy image sets with different scenes so as to extract the prior of GPDR, and the correctness of gray difference-ratio prior was verified based on the degradation model of visual field image. Secondly, the Naka-Rushton fitting function was set up according to constraint prior, and the parameter (n,k) of standard image sets from fitting NR function were calculated, and a lookup table of (n,k) corresponding to the fog concentration in the field of view were established. Thirdly, regression analysis was used to calculate the fitting parameter (n",k") of scatter plot of real foggy image, and the comprehensive correlation coefficient between the parameter (n,k) of the standard image sets and the parameter (n",k") of real image sets were calculated, and the parameter (n,k) whose correlation coefficient exceeded the threshold through searching the lookup table were concerned as the level valuation of concentration inspection. Simulations show that result of this algorithm is in line with the concentration change trend by test for fog image with different concentrations, and simulations also show that correlation coefficients between the results of this paper with PM2.5 can be up to 0.95 by test of samples with different concentrations both in the same scenes and in different scenes. This shows that the algorithm can be used as the fog concentration rating for visual field, and the horizontal comparison test shows that inspection accuracy of the algorithm in this paper can be up to 4.8%.

TP391.9

国家自然科学基金(52172324)；陕西省科技攻关项目(2015GY052)；陕西省交通厅重点项目(20-38T).西安市未央区科技计划(202121)；