基于码书的雾天图像质量评价方法研究

Research on Foggy Image Quality Assessment Based on Codebook

近年来,无参考图像质量评价发展迅速,但是对雾天图像质量进行评价的无参考算法还鲜有报道。该文提出了一种基于码书的无参考雾天图像质量评价算法。目的是使该方法评价雾天图像质量的结果与人类主观感知相一致。寻找能反映雾天图像质量的特征,运用这些特征构建码书,然后用码书对训练图像进行编码得到训练图像的特征向量,最后用这些向量与训练图像的主观评分进行回归得到雾天图像质量评价模型。该方法在仿真的雾天图像库中进行了测试,结果表明:Pearson线性相关系数和Spearman等级相关系数值都在0.99以上。并与经典的无参考算法NIQE和CONIA方法进行了比较,优于这些算法,能够很好地预测人对雾天图像的主观感知。

In recent years, no-reference image quality assessment has developed rapidly. However, algorithm about foggy image quality assessment has nearly reported. Our paper proposes an algorithm of foggy image quality assessment based on codebook. The goal of our paper is made our assessment consequence consistence with human opinion scores. Our method is to search the feature that can reflect foggy image quality, structure codebook using this feature, acquiring feature vector by encoding the training images using the codebook, the last, going the regression between the feature vector and the human opinion scores of the training images. Our algorithm have tested on foggy image database, the result shows that PLCC and SROCC both exceed 0.99 and is better than no-reference image quality assessment algorithm of NIQE and CONIA, our algorithm can predict perception of foggy image.