No-reference image quality assessment for photographic images based on robust statistics

No-reference image quality assessment (NR-IQA) is developing rapidly, but there lacks of research on exploring robust statistics to improve the prediction accuracy and monotonicity of NR-IQA algorithms, in particular for assessing photographic images captured by different digital cameras where a variety of unknown distortions may happen. Hence this paper proposes a novel robust-statistics-based NR-IQA method (termed RSN) for photographic images. In RSN, we present three types of features based on robust statistics: robust natural scene statistics of multiple components, robust multi-order derivatives, and robust complementary features in the frequency domain. Then support vector regression is applied to predict image quality using the extracted features. Experimental results show that RSN remarkably outperforms state-of-the-art NR-IQA methods on the CID2013 database of photographic images, as well as on the popular LIVE and TID2013 databases.