No-Reference Screen Content Image Quality Assessment With Unsupervised Domain Adaptation

In this paper, we quest the capability of transferring quality natural scene images to that are not acquired by optical cameras (e.g., screen content images, SCIs), rooted in widely accepted view human visual system has adapted and evolved through perception environment. Here, develop first unsupervised domain adaptation based no reference assessment method for SCIs, leveraging rich subjective ratings (NIs). general, it is a non-trivial task directly transfer prediction model from NIs new type (i.e., SCIs) holds dramatically different statistical characteristics. Inspired transferability pair-wise relationship, proposed measure operates on philosophy improving discriminability simultaneously. particular, introduce three types losses which complementarily explicitly regularize feature space ranking progressive manner. Regarding discriminatory enhancement, propose center loss rectify classifier improve its only source (NI) but also target (SCI). For discrepancy minimization, maximum mean (MMD) imposed extracted features SCIs. Furthermore, further enhance diversity, correlation penalization between dimensions, leading with lower rank higher diversity. Experiments show our can achieve performance source-target settings light-weight convolution neural network. The sheds light learning measures unseen application-specific without cumbersome costing evaluations.