SIQD: Surveillance Image Quality Database and Performance Evaluation for Objective Algorithms
Wenhan Zhu, Guangtao Zhai, Chen Yao, Xiaokang Yang · 2018
The surveillance camera is an important security device for police to maintain public order and provides clues to trace suspects. The quality of surveillance videos/images may be degraded during acquisition, compression and communication. It is difficult to acquire key information, such as human faces, with the bad quality surveillance videos/images. So the research of the quality assessment of surveillance images is quite necessary. In this paper, we perform a study on subjective quality evaluation of surveillance images and investigate whether the existing objective quality measures can be applied to the surveillance images. Concretely, we establish a new surveillance image quality database (SIQD) including 500 surveillance images with different degrees of quality through subjective study. Next, we investigate the prediction performance of the existing popular image quality assessment (IQA) algorithms on the SIQD database, which include eleven NR methods and four sharpness models. Experimental results demonstrate that the present objective models do not work well and quality measures having high correlation with human visual perception are high needed. Some meaningful findings are also given, which may enlighten the design of objective surveillance IQA models. The SIQD will be made publicly available to facilitate further surveillance IQA researches.