Large-Scale Works on Algorithms of Quality of Experience (QoE)

Mikoaj Leszczuk, Piotr Romaniak, Andrzej Gowacz, Jan Derkacz, Andrzej Dziech Academic Computer Centre CYFRONET {leszczuk,romaniak,glowacz,derkacz,Dziech}@kt. agh.edu.pl

Introduction and Motivation

Large variety of multimedia data sources means the variability not only in terms of included content, but also presented in terms of quality. This applies in particular to video sequences, where does exists the extremely variable quality of e.g. the present recordings made with amateur camera and e.g. recordings of specialized medical tests used for diagnostic purposes. Quality assessment provides therefore an extra, new dimension. In terms of mechanisms to deliver high-quality digital content, a common feature of existing systems is the need to ensure acceptable quality of streaming video sequences, regardless of the load transmission medium, the type of access network or terminal equipment which uses the end user. The term acceptable quality is not clearly defined, and mostly depends on the scenario under consideration. Video streaming system that is not able to provide acceptable quality ceases to be attractive to potential users and thus in general it is not in use. Due to the above requirements, an important element of the research is to develop a system used to ensure adequate quality of video sequences. The system would include metrics for quality assessment and quality optimization mechanisms that use information provided by the metrics.

The SYNAT Project

One of the aims of research within the SYNAT project is developing methods of delivering highquality multimedia content. It is planned to implement a variety of methods for segmentation of digital objects, transcoding and adaptation of multimedia content. This will include metrics and algorithms for assessing the quality of video sequences. It is planned to store additional information about the quality of digital objects for the materials that meet the criteria.

Novelty
The main innovation of the proposed solutions is a comprehensive evaluation of the perceived quality of video and optimization based on information provided by the proposed metrics. The complexity of the evaluation is related to the fact that it takes both the distortions typical to the video acquisition process and those caused by lossy compression for streaming. Through subjective testing and application of advanced statistical methods for the analysis of data received, it will be possible to create models mapping measured quality parameters onto values of the resultant video 1/2

quality, expressed in terms understandable to the user level (e.g. 5-point scale of the Mean Opinion Score, MOS). Another important issue is to optimize the quality of the information provided by the metric, written in the form of meta-data, or computed on demand. Optimizing the quality will take into account not only the network parameters (e.g. available bandwidth) but also the characteristics of video sequences, which is an important innovation.

Implementation
The work on algorithms of Quality of Experience (QoE) in the SYNAT Project were synchronized with those of the Video Quality Experts Group (VQEG), in particular the project of Joint Efforts Group (JEG) Hybrid group. The objective of JEG-Hybrid group is to assess the quality of video sequences based on the bit stream and decoded frames and motion, that is, information that is readily available on the DVD player/receiver. The first assessment of JEG-Hybrid group is limited to video encoding based on H.264 standard and compatible with the packet transmission systems such as IP networks. The recent activity in the Project SYNAT, made, in sync with JEG-Hybrid, the choice sets of Source Video Sequences (SRCs), and the selection of parameter sets Hypothetical Reference Circuits (HRCs).

Collections of Source Video Sequences (SRCs)

In this experiment, it was assumed that there are no restrictions imposed on the overall quality of the SRC video. However, the content of the SRC, which is worse in quality than the result of MOS value of 4, is to be treated separately. For example, in measurements in a model Full Reference, if a reliable prediction of subjective assessment is wanted, this type of content cannot be used. This is a preliminary estimate. Some video sequences can be repeated in some collections, which will require further verification. Many sequences origin from the database of Consumer Digital Video Library (CDVL).

Hypothetical Reference Circuit (HRC) Parameter Sets

The set of Hypothetical Reference Circuit (HRC) parameters defines the parameter set for a video encoder, which can be used for compression. Each of the proposed HRCs differs from the others in at least one of the considered parameters. The entire range of HRCs should cover all scenarios of probable video compression, designed for digital storage in digital libraries. Each Source Video Sequence (SRC) should be distorted (encoded) using each of the HRCs, which results in a number of distorted Processed Video Sequences (PVSs) equal to SRC x HRC sequences. Video compression parameters considered are the following: Constant bitrate of at least 5 levels Constant quantization factor of at least 5 levels Group of Pictures (GoP) structure different sizes and structures: o Of at least 2 levels with a variable number of I frames per second o Of at least 2 levels with a variable number of B and P frames per second o Hierarchical coding of at least 2 horizontal variants Structure of slices of at least 2 levels Number of frames per second of at least 2 levels (original and halved) Resolution of at least 2 levels (original and halved) 2/2