Normalization of subjective video test results using a reference test and anchor conditions for efficient model development

A set of five video subjective tests have been conducted for developing a video quality model. In this paper, we describe the prerequisites for merging the subjective tests results. In particular, we detail the use of a reference test and anchor conditions for normalizing the ratings. The reference test is designed and the anchor conditions selected so that they span a wide range of quality and contain all presumed perceptual dimensions present in the application domain of the model (IPTV). The anchor conditions are then used for normalizing the results of the tests, i.e. for mapping them to the ones of the reference test. Some non-anchor conditions are identical between at least two tests. The impact of using anchor conditions is thus evaluated by comparing the ratings of those identical conditions in terms of Pearson correlation and Root Mean Square Error. Since we used the test results for modeling video quality, the impact of using anchor conditions is further examined in terms of modeling performance.