Towards Predictions of the Image Quality of Experience for Augmented Reality Scenarios

Abstract: Augmented Reality (AR) devices are commonly head-worn to overlay context-dependent information into the field of view of the device operators. One particular scenario is the overlay of still images, either in a traditional fashion, or as spherical, i.e., immersive, content. For both media types, we evaluate the interplay of user ratings as Quality of Experience (QoE) with (i) the non-referential BRISQUE objective image quality metric and (ii) human subject dry electrode EEG signals gathered with a commercial device. Additionally, we employ basic machine learning approaches to assess the possibility of QoE predictions based on rudimentary subject data. Corroborating prior research for the overall scenario, we find strong correlations for both approaches with user ratings as Mean Opinion Scores, which we consider as QoE metric. In prediction scenarios based on data subsets, we find good performance for the objective metric as well as the EEG-based approach. While the objective metric can yield high QoE prediction accuracies overall, it is limited i its application for individual subjects. The subject-based EEG approach, on the other hand, enables good predictability of the QoE for both media types, but with better performance for regular content. Our results can be employed in practical scenarios by content and network service providers to optimize the user experience in augmented reality scenarios.