No-Reference Quality Assessment of Night-Time Images Via the Analysis of Local and Global Features

This paper focuses on the quality assessment of the night-time images, which is technically vulnerable to diverse and sophisticated visual distortions, and proposes an effective no-reference image quality assessment method through comprehensive analysis of local and global image attributes. Several quality-aware features are extracted through the study of image brightness, contrast, structure and color. Specifically, the features related to brightness and contrast are extracted through the analysis of local information, while the others are extracted through the analysis of global information. The quality assessment model for night-time images is built by mapping the selected features into subjective ratings via the support vector regression. Extensive experiments conducted on public night-time image database demonstrates the superiority of the proposed method.