Towards still image experience predictions in augmented vision settings

With the emergence of Augmented Reality (AR) services in a broad range of application scenarios, the interplay of (compressed) content as delivered and quality as experienced by the user becomes increasingly important. While significant research efforts have uncovered interplays for traditional (opaque) media consumption scenarios, applications in augmented vision scenarios pose significant challenges. This paper focuses on the Quality of Experience (QoE) for a grounded truth reference still image set. We corroborate previous findings which indicate that there is only limited QoE benefit obtainable from very high image quality levels. The main contribution is the first evaluation of several popular objective image quality metrics and their relationships to the QoE in opaque and vision augmenting presentation formats. For the first time, we provide an assessment of the possibility to predict the QoE based on these metrics. We find that linear regression of a small degree is able to accurately capture the coefficients and fairly accurate prediction can be performed even with a non-referential image quality metric as parameter. We note, however, that prediction accuracy still fluctuates with the image content and subjects removed from the data set. Our overall findings can be employed by AR service providers to perform an optimized delivery of still image content.