基于自然统计特征分布的交通图像雾浓度检测

针对交通场景图像去雾中缺乏有效浓度检测的不足，提出基于自然统计特征分布的雾浓度检测算法.首先，将待检图像分为P×P大小的子块；其次，建立图像局部对比度及熵的自然统计特性向量，求解待检图像和标准图像子集间最佳期望及协方差的最大似然估计；最后，分别计算待检图像与有雾和无雾标准子集间的马氏距离，以二者的比值D作为场景雾浓度的度量.通过同一场景不同浓度等级和不同场景不同雾浓度等级的图像测试，表明算法D值能正确反应雾浓度的变化趋势；通过与标准主观评价法（MOS）比较，表明二者呈现近似的线性，相关系数可达0.97，远高于归一化亮度系数（MSCN）的0.56；通过与PM2.5比较，表明算法能准确的评定雾浓度等级.

Inspection of Fog Density for Traffic Image Based on Distribution Characteristics of Natural Statistics

Concerning low efficiency detection for foggy density in traffic image,a novel algorithm was proposed to check foggy density based on distribution characteristics of natural statistics.Firstly,foggy images were partitioned as P×P pixel patches by method of the maximum overlap count.Secondly,featured function vector for local contrast and entropy was created and maximum likelihood estimation between tested image and two standard image corpuses were respectively computed.Finally,Mahalanobis-like Distances(MD) between foggy image and corpus of standard foggy image or fog-free image were achieved,and the ratio of two values could be used to measure the foggy density.Simulation shows that the value D can respond the varied tendency to density for same scene with difference density or different scene with different density.Correlation coefficient up to 0.97 between this algorithm and mean opinion scores (MOS) method indicate high linear about them and the coefficient is larger than 0.56 between mean subtracted contrast normalized (MSCN) and MOS.Comparison to PM2.5 shows that this algorithm can be used to evaluate the level for fog density.