Subjective Quality Assessment of H.264/AVC Video Streaming with Packet Losses
© Francesca De Simone et al. 2011
Received: 15 November 2010
Accepted: 18 January 2011
Published: 27 January 2011
Research in the field of video quality assessment relies on the availability of subjective scores, collected by means of experiments in which groups of people are asked to rate the quality of video sequences. The availability of subjective scores is fundamental to enable validation and comparative benchmarking of the objective algorithms that try to predict human perception of video quality by automatically analyzing the video sequences, in a way to support reproducible and reliable research results. In this paper, a publicly available database of subjective quality scores and corrupted video sequences is described. The scores refer to 156 sequences at CIF and 4CIF spatial resolutions, encoded with H.264/AVC and corrupted by simulating the transmission over an error-prone network. The subjective evaluation has been performed by 40 subjects at the premises of two academic institutions, in standard-compliant controlled environments. In order to support reproducible research in the field of full-reference, reduced-reference, and no-reference video quality assessment algorithms, both the uncompressed files and the H.264/AVC bitstreams, as well as the packet loss patterns, have been made available to the research community.
The use of IP networks for video delivery is gaining an increasing popularity as a mean of broadcasting data from content providers to consumers. Video transmission over peer-to-peer networks is also becoming very popular. Typically, these networks provide only best-effort services, that is, there is no guarantee that the content will be delivered without errors. In practice, the received video sequence may be a degraded version of the original. Besides distortions introduced by lossy coding, user's experience might be also affected by channel-induced distortions. Thus, the design of systems for automatic monitoring of the received video quality is of great interest for service providers, in order to optimize the transmission strategies as well as to ensure a desired level of quality of experience. Several algorithms, usually referred to as objective quality metrics, have been proposed in the literature for the in-service objective quality evaluation of video sequences. They consist of No-Reference or Reduced-Reference methods relying on the analysis of the bitstream, the pixels, or both (so-called hybrid approach) [1, 2]. Nevertheless, the lack of publicly available databases of video sequences and subjective scores makes the comparison of existing and novel solutions very difficult.
In fact, research in the field of video quality assessment relies on the availability of subjective scores, collected by means of experiments in which groups of people are asked to rate the quality of video sequences. In order to gather reliable and statistically significant data, subjective tests have to be carefully designed and performed and require a large number of subjects. For these reasons, the subjective tests are usually very time consuming. Nevertheless, the availability of subjective scores is fundamental to enable validation and comparative benchmarking of objective video quality metrics in a way to support reproducible and reliable research results.
The first public database of video contents and related subjective quality scores was produced by the Video Quality Experts Group (VQEG) and used to compare the performance of Full-Reference objective metrics, targeting secondary distribution of television as application . Unfortunately, only part of the subjective results and test materials used to perform the study has been made publicly available. Additionally, this dataset includes interlaced video sequences and focuses on MPEG-2 compression. These distortions are not representative of the current video coding and transmission technologies. Thus, the usage of VQEG data by independent researchers to validate more recent and future metrics is limited. Recently, two video databases have been proposed by the Laboratory for Image and Video Engineering (LIVE) at the University of Texas at Austin. The LIVE Video Quality Database  consists of a set of video sequences corresponding to different contents, distorted by MPEG-2 and H.264/AVC compression as well as by transmission over error-prone IP and wireless networks. The presence of diverse distortion types makes this database particularly useful to test the consistency of metrics performance. The LIVE Wireless Video Quality Assessment Database  focuses on distortions due to transmission over a wireless network and takes into account a set of video sequences having similar content concerning airplanes. These databases include the test video sequences and the processed subjective results and have been used to evaluate the performance of a set of Full-Reference video quality metrics in [4, 5].
In this paper, a detailed description of the publicly available database originally presented in  and extended in  is provided, along with an extensive discussion of the data processing applied to the collected subjective scores and analysis of the results. The database focuses on the impact of packet losses on visual quality. It contains subjective scores collected through subjective tests carried out at the premises of two academic institutions: Ecole Polytechnique Fédérale de Lausanne-Switzerland and Politecnico di Milano-Italy. The same experiments were performed at both laboratories and a total of 40 subjects were asked to rate 144 video sequences, corresponding to 12 different video contents at CIF and 4CIF spatial resolutions and different Packet Loss Rates (PLRs), ranging from 0.1% to 10%. The packet loss free sequences were also included in the test material, thus in total 156 sequences were rated by each subject at each institution.
With respect to others cited above, the database described in this paper, (1) includes data collected at the premises of two different laboratories, showing high correlation among the two sets of collected results, as an indicator of reliability of the subjective data as well as of the adopted evaluation methodology, (2) includes the decoded sequences and the compressed video streams affected by packet losses, as well as the packet loss patterns, thus it can be used for testing stream-based and hybrid No-Reference and Reduced-Reference metrics, (3) includes the complete set of collected subjective results, including the raw scores before any data processing, thus allowing reproducible research on subjective data processing and detailed statistical analysis of metrics performance. The database is available for download at http://mmspg.epfl.ch/vqa and http://vqa.como.polimi.it.
The rest of the paper is organized as follows. Section 2 describes the test material, the environmental setup, and the subjective evaluation methodology used in our study. In Section 3, the processing of the results is detailed. The results of the two laboratories are analyzed and compared in Section 4. Finally, Section 5 concludes the paper.
2. Subjective Video Quality Assessment
In a subjective video quality test, a group of people is asked to watch a set of video sequences and to rate their quality. The design of formal subjective experiments involves four main phases [8–10]:
(1) Selection of the Test Material
The test material has to be a realistic sample of the actual data that belongs to the target application scenario. Also, in order to avoid decreasing subject's level of attention, the content has to be heterogeneous and the test sessions should not last more than 30 minutes each, including any training phase. For the same reason, it is important to select stimuli whose quality levels are possibly uniformly distributed across the rating scale. Therefore, an accurate supervised screening of the test material is needed. Whenever it is not possible to show the entire set of the test materials in a single test session (for instance, because the duration of the session exceeds 30 minutes), multiple sessions may be scheduled, so that each subject is able to perform all the sessions and rate all the test material (i.e., full factorial design). Alternatively, a reduced subset of test conditions may be selected.
(2) Selection of the Test Methodology
Several internationally accepted test methods for subjective video quality assessment are described in [11, 12]. A first taxonomy of the methods regards how the visual stimuli are presented to the viewer. In Double Stimulus methods, the observer is sequentially presented with two video sequences: one of the two sequences is the reference stimulus and the other is the test stimulus. The observer can be asked to rate either both stimuli, or only the test stimulus. In Single Stimulus methods, only one stimulus is shown and has to be rated. Finally, in Stimulus Comparison methods, pairs of stimuli are shown simultaneously and the subject is asked to compare their quality. A second classification of test methodologies concerns the rating scale in which the subject is asked to express her/his quality evaluation score. A first distinction is between continuous and discrete scales. A second distinction pertains the use of either a categorical scale (textual labels, describing the quality of the stimulus or the annoyance of the impairments) or a numerical scale. Finally, the subject may be asked to enter her/his rating after the visualization of the test material, and/or directly while watching the video sequence. Such continuous evaluations can be used to elicit an indication of the temporal quality variations across the sequence.
(3) Selection of the Participants
In order to gather statistically significant data, subjective tests require a large enough number of subjects, as a representative sample of the population of interest. The participants to the test have to be screened for visual acuity and color blindness. They can be chosen from two categories of end users depending on the goal of the investigation: expert or naive, that is, nonexpert.
(4) Choice of the Experimental Setup
The experimental setup should reproduce the viewing conditions of the target application scenario, while keeping under control all the external experimental parameters which could influence subject's perception. Some recommendations and setup parameters are indicated in [11, 13]. An accurate control of the test environment is necessary to ensure the reproducibility of the test activity and to compare results across different laboratories and test sessions.In the following, the test material, the environment setup, the subjective evaluation methodology, and the panel of subjects used in our study are described.
2.1. Test Material
H.264/AVC encoding parameters.
Number of frames
4 : 2 : 0
Number of reference frames
Fixed number of macroblocks
Disabled, fixed QP (Table 2)
Macroblock partitioning for motion estimation
Motion estimation algorithm
Enhanced Predictive Zonal Search (EPZS)
Early skip detection
Selective intramode decision
Test sequences and coding conditions.
Spatial res. and fps
CIF 30 fps
CIF 30 fps
CIF 30 fps
CIF 30 fps
CIF 30 fps
CIF 30 fps
4CIF 30 fps
4CIF 30 fps
4CIF 30 fps
4CIF 25 fps
4CIF 25 fps
4CIF 25 fps
For each of the twelve original H.264/AVC bitstreams, a number of corrupted bitstreams were generated, by dropping packets according to a given error pattern . Coded slices belonging to the first frames were not corrupted, as they contained header information (Picture Parameter Set (PPS) and Sequence Parameter Set (SPS)). Conversely, the remaining slices might be discarded from the coded bitstream. To simulate burst errors, the patterns were generated at six different PLRs, 0.1%, 0.4%, 1%, 3%, 5%, 10%, with a two-state Gilbert's model . The model parameters were tuned to obtain an average burst length of 3 packets, which is a typical characteristic of IP networks . The two-state Gilbert's model generated, for each PLR, several error patterns. For each PLR and content, two decoded video sequences were manually selected in order to uniformly span a wide range of distortions, that is, perceived video quality, while keeping the size of the dataset manageable. The details of the selection procedure can be found in . A total of 72 CIF sequences with packet losses and 72 4CIF sequences with packet losses were included in the test material. Each bitstream was decoded with the H.264/AVC reference software decoder with motion-compensated error concealment turned on .
2.2. Environment Setup
Specifications of LCD display devices.
Samsung SyncMaster 920N
2560 × 1600 (native)
1280 × 1024 (native)
EyeOne Display 2
EyeOne Display 2
2.3. Test Methodology
The Single Stimulus (SS) method was used to collect the subjective data. Thus, each processed video sequence was presented alone, without being paired with its unprocessed, that is, reference, version. However, the test procedure included a reference version of each video sequence, which in this case was the packet loss free sequence, as a freestanding stimulus for rating like any other. At the end of each test presentation, a voting time followed, when the subjects were asked to rate the quality of the stimulus using a five-point ITU continuous adjectival scale.
Each test session referred to a single spatial resolution (i.e., either CIF or 4CIF) and included 83 video sequences: 6 × 12 test sequences, that is, realizations corresponding to 6 different contents and 6 different PLRs; 6 reference sequences, that is, packet loss free video sequences; 5 stabilizing sequences, that is, dummy presentations shown at the beginning of the experiment to stabilize observers' opinion. The dummy presentations consisted in 5 realizations, corresponding to 5 different quality levels, selected from the test video sequences. The results for these items were not registered by the evaluation software but the subject was not told about this. The presentation order for each subject was randomized, discarding those permutations where stimuli related to the same original content were consecutive.
Before each test session, written instructions were provided to subjects to explain their task. Additionally, a training session was performed to allow viewers to familiarize with the assessment procedure and the software user interface. The contents shown in the training session were not used in the test sessions and the data gathered during the training was not included in the final test results. In particular, for the training phase two different contents for each spatial resolution and five realizations of each, representatives of the score labels depicted in Figure 4, were used. During the display of each training sequence, the operator explained the meaning of each label, as summarized in the written instructions reported in Appendix A.
The entire set of test sessions was performed in each laboratory. Twenty-three and twenty-one subjects took part in the CIF and 4CIF sessions, respectively, at PoliMI. Seventeen and nineteen subjects took part in the CIF and 4CIF sessions, respectively, at EPFL. All subjects reported that they had normal or corrected to normal vision. Their age ranged from 24 to 40 years old. All observers were naive subjects.
3. Subjective Data Processing
Some general guidelines for processing the results of quality assessment experiments are detailed in  but, when dealing with subjective data, the statistical tools to be used need to be selected according to the properties of the data under analysis. Thus, a case-by-case approach is preferable.
In general, the results of a subjective experiment for quality assessment are summarized by averaging the scores assigned by the panel of observers to each video sequence, that is, stimulus, in order to obtain a Mean Opinion Score (MOS) and corresponding Confidence Interval (CI) . Prior to the MOS computation, a correction procedure, that is, normalization, to compensate for any systematic differences in the usage of the rating scale by the different subjects can be applied. Then, the scores are screened in order to detect and exclude possible outliers, that is, subjects whose scoring significantly deviates from others.
The scores collected in the CIF and 4CIF sessions by the two laboratories were processed separately, according to the procedure detailed below.
3.1. Scores Normalization
First, in order to check the between-subject variability, a two-way ANOVA was applied to the raw scores . The results of the ANOVA showed that a subject-to-subject correction was needed. Thus the scores were normalized according to offset mean correction rule. The details of the normalization procedure are described in Appendix B. This approach differs from that applied in [4, 5, 21], where the normalization procedure consists in converting the scores to z-scores, mainly because not only the mean score of the single subject is taken into account to correct his/her own scores, but also the overall mean across all test conditions and subjects.
3.2. Subjects Screening
The screening of possible outlier subjects was performed considering the normalized scores, according to the procedure described in Appendix C. Three and one outliers were detected out of 23 and 17 subjects, from the results produced for CIF data at PoliMI and at EPFL, respectively. Four and two outliers were detected out of 21 and 19 subjects, from the results produced for 4CIF data at PoliMI and at EPFL, respectively. The scores corresponding to the outlier subjects were discarded from the results.
3.3. Mean Opinion Scores and Confidence Intervals
where is the -value associated with the desired significance level for a two-tailed test ( ) with degrees of freedom, where denotes the number of observations in the sample (i.e., the number of subjects after outliers detection) and the estimated standard deviation of the sample of observations.
4. Analysis of the Results
Additionally a Hotellings -test for two series of population means  has been applied, separately to the results of the CIF and the 4CIF experiments, to understand whether the data of the two laboratories could be merged to compute overall MOS and associated CI values to be used as benchmark values for example for testing the performance of objective metrics. The null hypothesis is of no difference between the two multivariate patterns of scores, that is the subjects in the two laboratories do not differ in their responses to the stimuli. The result of the test for both the resolutions indicates that the the null hypothesis cannot be rejected, as a further proof of the fact that the two datasets of results could be merged for future studies involving the subjective results.
Finally, it is worth mentioning that the experiment for the quality assessment of CIF sequences was also carried out by performing the same evaluation under uncontrolled environmental conditions, in order to analyze the effect of the environment on the subjective scoring. A laptop was used to show the GUI and the test room was a normal office or living room. Twelve subjects took part in the experiments. Unfortunately, the results do not allow drawing any conclusion upon a systematic effect of the environmental conditions on the scores, since according to the content under analysis a different behavior of the MOS values, with respect to the results of the formal evaluations, was noticed. Currently an investigation is in progress in order to better understand the mechanisms behind the variability of the obtained results.
5. Concluding Remarks
In this paper, a database, containing the results of a subjective evaluation campaign aiming at studying the subjective quality of video sequences affected by packet losses, is presented. Subjective data was collected at the premises of two institutions, Ecole Polytechnique Fédérale de Lausanne and Politecnico di Milano. The database is publicly available at http://mmspg.epfl.ch/vqa and http://vqa.como.polimi.it and contains data relative to 156 sequences, both at CIF and 4CIF spatial resolutions. More specifically, the following data are provided: (1) the test material, together with the software tools used to produce them, (2) the corresponding H.264/AVC bitstreams, useful for evaluating no-reference and reduced-reference metrics, (3) the raw subjective scores, (4) the final MOS and CI data, together with the Matlab implementation of the algorithms adopted for score normalization and outlier screening.
The results of the subjective tests performed in two different laboratories show high consistency and correlation. We believe that such a publicly available database will allow easier comparison and performance evaluation of the existing and future objective metrics for quality evaluation of video sequences, contributing to the advance of the research in the field of objective quality assessment.
Future works will focus on an extension of the current database in order to include distortions due to jitter and delay. Also, further investigation on the effect of the environmental conditions over the results of the subjective experiments will be performed, focusing on the mobile scenario, where the assessment of video quality for test material at CIF spatial resolution is more realistic.
This work has been partially sponsored by the EU under PetaMedia Network of Excellence and by the Swiss National Foundation for Scientific Research in the framework of NCCR Interactive Multimodal Information Management (IM2). Part of the material presented in this paper has been published in [6, 7].
- Hemami SS, Reibman AR: No-reference image and video quality estimation: applications and human-motivated design. Signal Processing 2010, 25(7):469-481.Google Scholar
- Winkler S, Mohandas P: The evolution of video quality measurement: from PSNR to hybrid metrics. IEEE Transactions on Broadcasting 2008, 54(3):660-668.View ArticleGoogle Scholar
- Corriveau P, Webster A: Final report from the video quality experts group on the validation of objective models of video quality assessment, phase II. Video Quality Expert Group; 2003.Google Scholar
- Seshadrinathan K, Soundararajan R, Bovik AC, Cormack LK: Study of subjective and objective quality assessment of video. IEEE Transactions on Image Processing 2010, 19(6):1427-1441.View ArticleMathSciNetGoogle Scholar
- Moorthy AK, Seshadrinathan K, Soundararajan R, Bovik AC: Wireless video quality assessment: a study of subjective scores and objective algorithms. IEEE Transactions on Circuits and Systems for Video Technology 2010, 20(4):587-599.View ArticleGoogle Scholar
- De Simone F, Naccari M, Tagliasacchi M, Dufaux F, Tubaro S, Ebrahimi T: Subjective assessment of H.264/AVC video sequences transmitted over a noisy channel. Proceedings of the International Workshop on Quality of Multimedia Experience (QoMEx '09), 2009 204-209.Google Scholar
- De Simone F, Tagliasacchi M, Naccari M, Tubaro S, Ebrahimi T: A H.264/AVC video database for the evaluation of quality metrics. Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP '10), 2010, Dallas, Tex, USA 2430-2433.Google Scholar
- Wu HR, Rao KR: Digital Video Image Quality and Perceptual Coding (Signal Processing and Communications). CRC Press; 2005.View ArticleGoogle Scholar
- Winkler S: Digital Video Quality—Vision Model and Metrics. John Wiley & Sons, New York, NY, USA; 2005.View ArticleGoogle Scholar
- Bech S, Zacharov N: Perceptual Audio Evaluation: Theory, Method, and Application. John Wiley & Sons, New York, NY, USA; 2006.View ArticleGoogle Scholar
- ITU-T : Recommendation ITU-R BT 500-10. Methodology for the subjective assessment of the quality of the television pictures, 2000Google Scholar
- ITU-T : Recommendation ITU-R P 910. Subjective video quality assessment methods for multimedia applications, 1999Google Scholar
- VQEG hybrid testplan, version 1.2, ftp://vqeg.its.bldrdoc.gov
- Joint Video Team (JVT) : H.264/AVC reference software version JM14.2. http://iphome.hhi.de/suehring/tml/download/
- Luttrell M, Wenger S, Gallant M: New versions of packet loss environment and pseudomux tools. Joint Video Team (JVT); 1999.Google Scholar
- Gilbert EN: Capacity of a burst-noise channel. Bell System Technical Journal 1960, 39: 1253-1266.View ArticleMathSciNetGoogle Scholar
- Chua TK, Pheanis DC: QoS evaluation of sender-based loss-recovery techniques for VoIP. IEEE Network 2006, 20(6):14-22.View ArticleGoogle Scholar
- Sullivan GJ, Wiegand T, Lim K-P: Joint model reference encoding methods and decoding concealment methods. Joint Video Team (JVT); 2003. JVT-I049Google Scholar
- Mplayer, http://www.mplayerhq.hu/
- Snedecor GW, Cochran WG: Statistical Methods. Iowa State University Press; 1989.MATHGoogle Scholar
- Sheikh HR, Sabir MF, Bovik AC: A statistical evaluation of recent full reference image quality assessment algorithms. IEEE Transactions on Image Processing 2006, 15(11):3440-3451.View ArticleGoogle Scholar
- Mardia KV, Kent JT, Bibby JM: Multivariate Analysis. Academic Press; 1979.MATHGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.