Metrics for Video Quality Assessment in Mobile Scenarios

TL;DR

This paper evaluates various video quality metrics in LTE A mobile networks, identifying the most effective no-reference metric for assessing video quality under different network conditions.

Contribution

It characterizes the performance of popular video quality metrics in LTE A networks and identifies BRISQUE as the most effective no-reference metric for mobile scenarios.

Findings

Blocking metric is effective for channel or modulation variations.

BRISQUE performs well in quantizing compression and network distortions.

Different metrics have varying strengths depending on the type of network variation.

Abstract

With exponential increase in the volumes of video traffic in cellular net-works, there is an increasing need for optimizing the quality of video delivery. 4G networks (Long Term Evolution Advanced or LTE A) are being introduced in many countries worldwide, which allow a downlink speed of upto 1 Gbps and uplink of 100 Mbps over a single base station. This makes a strong push towards video broadcasting over LTE networks, characterizing its performance and developing metrics which can be deployed to provide user feedback of video quality and feed-back them to network operators to fine tune the network. In this paper, we characterize the performance of video transmission over LTE A physical layer using popular video quality metrics such as SSIM, Blocking, Blurring, NIQE and BRISQUE. We conduct experiments to find a suitable no-reference metrics for mobile scenario and find that Blocking Metrics is most promising in case of channel or modulation variations but it does not perform well to quantize variations in compression ratios. The metrics BRISQUE is very efficient in quantizing this distortion and performs well in case of network variations also.