Image Quality Assessment for Video Surveillance System

With the popularity of surveillance system, traditional method to daily keep watch on its performance by human cannot meet the requirements anymore. Image degradation is a progressive process and its ideal version can be captured at beginning. The objects in the scene may change during its usage, so that the image content to be examined will be significantly different with the referred one. Therefore, the full-reference (FR) image quality assessments (IQAs) are no longer efficient for this application. In this paper, a reduced-reference (RR) IQA is proposed to fit the distribution of MSCN coefficients as the low level feature, and the feature is combined with the content representation. This feature is associated with MOS by SVR to produce the IQA model. We validate the performance of our method with an extensive study involving 1000 surveillance images and experimental results show that the method fits with the subjective evaluation better than the existing FR and NR algorithms.