A Neural Network Based Test Bed for Evaluating the Quality of Video Streams in IP Networks

We introduce a modular methodology for evaluating the perceived quality of video streams in simulated packet networks through the use of artificial neural networks. One particular implementation of the test bed is presented and the results obtained with it under simple network configurations are discussed. Our tool was able to accurately predict the MOS scores of human viewers. Other applications of the test bed are also presented. For data analysis, we found that the usual parameters that can be controlled in an MPEG-4 codec do not have such a strong influence on the perceived video quality as a good network design that protects the video flows may do