Quality Assessment for Natural and Screen Content Images

Quality assessment (QA) of screen content images (SCIs) has gained more and more popularity. SCIs are very different from natural images (NIs) which have been dealing with by most researchers in the literature. QA methods specifically designed for NIs also can be used to evaluate the quality of SCIs. Yet, their performances are unsatisfactory. This may due to the statistical differences of SCIs and NIs. In this paper, SCIs and NIs QA methods in the literature are being compared and studied for both SCIs and NIs benchmarked databases. It is found out that methods that incorporate gradient features work well for both SCIs and NIs. This points out a possible way to utilize gradient features to come out with a QA method that works for both SCIs and NIs simultaneously. Hence, application related to SCIs and NIs such as deep learning and multitasking for person tracking system can be improved with the QA method.