DeSVQ: Deep Learning Based Streaming Video QoE Estimation
Monalisa Ghosh, Chetna Singhal, Rushikesh Wayal · 2022
The quality-of-experience (QoE) is a notable subjective quality metric to assess the efficiency of multimedia streaming services. Video streaming is a segment seeing notable growth in the past decade. It is affected by a mixed interplay of quality switching, compression, and buffering events that induce distortions that vary with time. Accurate streaming QoE prediction can help in adapting the content transmission to improve end viewers experience. In this work, we propose DeSVQ, a deep learning approach that uses an integrated framework consisting of Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM) networks that combine multiple feature processing stages, each of them effectively capturing the complex dependencies underlying the QoE prediction process. In one stage, the features are extracted per frame from the distorted videos by the CNN that are mapped sequentially to QoE scores by the LSTM network. In another stage, the LSTM network explores the temporal dependencies using the objective (numerical) features. The output from both the stages are linearly combined and fed to the decision trees. A cross-validation framework is used for evaluation. Our proposed model is shown to perform better QoE prediction over the existing approaches.