Subjective and Objective Quality Assessment of Compressed Screen Content Images

Objectively accessing the quality of screen content images (SCIs) is a challenging problem, as SCIs may not always have identical properties as natural scenes. Here we conduct comprehensive studies on the subjective and objective quality assessment of the compressed SCIs. Firstly, we build a database that contains the distorted SCIs generated by the high efficiency video coding standard as well as its extension on screen content compression. Subsequently, subjective experiments are conducted to evaluate the perceived quality of these SCIs with compression artifacts. To automatically predict the subjective quality, a reduced-reference quality assessment model is further learnt by a set of wavelet domain features concerning the generalized spectral behavior, the fluctuations of the energy, and the information content with relatively large scale training samples. Our experimental results show that the learnt model is able to achieve better prediction on the SCI quality with a few extracted meaningful features.