中文摘要：

针对监控图像受到多重噪声影响的特点,及安防对监控图像的信息量大小、子区域质量的特殊要求,提出一种基于空域特征的图像质量无参评估方法。该方法在图像自然场景统计模型空域特征基础上,引入图像二维信息熵作为特征之一;另外,提出分块分析方法,将图像的子区域质量并入图像特征;最后,通过机器学习优化特征权重,得到图像质量评价模型。交叉验证实验中,该方法对监控图像质量评估结果与主观质量得分的线性相关系数、斯皮尔曼等级相关系数均值分别为0.783和0.687,相对空域上的无参考质量评价方法 BRISQUE分别提高了0.7 d B和1.5 d B。实验结果表明,算法对监控图像质量评估结果与专业人士主观评价结果一致性明显高于对比算法。

英文摘要：

According to the characteristics that surveillance image is distorted by multiple noise,and quality of image corresponds with its amount of information as well as quality of sub-area,a no-reference surveillance image quality assessment method operated in the spatial domain was proposed. Based on natural scene statistics,2D entropy was introduced as its feature. Sub-areas＇ qualities wee also included by block analysis method. Finally,its parameters were optimized by machine learning to obtain scientific assessment model. In cross validation experiment,the correlation coefficient is 0. 783,which is0. 7 d B higher than that of blind / referenceless image spatial quality evaluator（ BRISQUE） and spearman rank-order correlation coefficient is 0. 687,which is 1. 5 d B higher than that of BRISQUE. The experiment shows that the result of our method to assess the quality of surveillance image is highly consistent to the subjective assessment by experts.