Constructing a No-Reference H.264/AVC Bitstream-Based Video Quality Metric Using Genetic Programming-Based Symbolic Regression

Abstract

Abstract-In order to ensure optimal quality of experience toward end users during video streaming, automatic video quality assessment becomes an important field-of-interest to video service providers. Objective video quality metrics try to estimate perceived quality with high accuracy and in an automated manner. In traditional approaches, these metrics model the complex properties of the human visual system. More recently, however, it has been shown that machine learning approaches can also yield competitive results. In this paper, we present a novel noreference bitstream-based objective video quality metric that is constructed by genetic programming-based symbolic regression. A key benefit of this approach is that it calculates reliable white-box models that allow us to determine the importance of the parameters. Additionally, these models can provide human insight into the underlying principles of subjective video quality assessment. Numerical results show that perceived quality can be modeled with high accuracy using only parameters extracted from the received video bitstream.Index Terms-H.264/AVC, high definition, no-reference, objective video quality metric, quality of experience (QoE).

Efficient lightweight video packet filtering for large-scale video data delivery

“…By using machine learning approaches, Staelens et al [17] extract useful information from the compressed video to estimate the perceived quality. The main goal of those studies is to be able to estimate the visibility of packet loss at the client to get the average perceived QoE.…”

Objective Video Quality Assessment — Towards Large Scale Video Database Enhanced Model Development

“…As a general term, Quality of Service (QoS), is used to denote the quality of the (delivery) network and is typically measured in terms of bandwidth, packet loss, delay, and jitter. A lot of research has already been conducted towards mapping QoS measurements to QoE prediction [20]- [23].…”

Supporting visual quality assessment with machine learning

“…Both feedforward neural networks and kernel machines -just as several other ML paradigms -involve the application of non-linear operators; as a result, they are often categorized as 'black box' approaches. In this regard, some recent work on video quality assessment [12] favored ML methodologies that could lead to 'white box' models, i.e., for which the predictive system can be translated into a set of intelligible rules. An interesting example of such methodologies is genetic programming (GP) [38,39].…”

“…Several studies proved the effectiveness of methodologies that exploit ML tools to address both video [5][6][7][8][9][10][11][12] and image [13][14][15][16][17][18][19][20][21][22][23][24] quality assessment. Furthermore, as a major confirmation of the potential of these technologies in perceptual quality assessment, a CI-based framework has been adopted in multiple methods for audio quality assessment, including the ITU standard, PEAQ [25].…”