Abstract:

This paper introduces an NR decision tree based QoE assessment model. The proposed model assesses quality of user experience of streaming videos using decision tree statistical learning method with a set of video-related features and network distortion features extracted from both the packet header at the packet level in the physical layer and at the video frame level in the application layer. These features are extracted solely from the packet header without further decoding of the video bitstreams, which decreases the computational complexity and makes the model independent of the encoding method. Thanks to decision tree’s high readability and fast classification speed, several decision trees have been built with different combinations of above feature subsets to study the relative importance of features. The result shows that the proposed model, which considers video-related features and both kinds of network distortions, outperforms the other resulting ones in terms of predict accuracy and monotonicity. This model can be used in the real-time streaming video quality monitoring systems.

Key words: QoE assessment, decision tree, video-related features, packet loss, delay