Abstract
Maximum Likelihood Difference Scaling (MLDS) used as a method for subjective assessment of video quality alleviates the inconveniencies associated with high variation and biases common in rating methods. However, the number of tests in a MLDS study rises fairly quickly with the number of samples that we want to test. This makes the MLDS studies not scalable for the diverse video delivery environments commonly met in pervasive media networks. To tackle this issue we have developed an active learning approach that decreases the number of MLDS tests and improves the scalability of this method. © 2012 ICST Institute for Computer Science, Social Informatics and Telecommunications Engineering.