Impact of Video Content on Video Quality for Video over Wireless Networks Asiya Khan, Lingfen Sun and Emmanuel Ifeachor Centre for Signal Processing and Multimedia Communication School of Computing, Communications and Electronics University of Plymouth, Plymouth PL4 8AA, UK. e-mail: (asiya.khan, l.sun, e.ifeachor)@plymouth.ac.uk Abstract—Video streaming is a promising multimedia application and is gaining popularity over wireless/mobile communications. The quality of the video depends heavily on the type of content. The aim of the paper is threefold. First, video sequences are classified into groups representing different content types using cluster analysis based on the spatial (edges) and temporal (movement) feature extraction. Second, we conducted experiments to investigate the impact of packet loss on video contents and hence find the threshold in terms of upper, medium and lower quality boundary at which users’ perception of service quality is acceptable. Finally, to identify the minimum send bitrate to meet Quality of Serive (QoS) requirements (e.g. to reach communication quality with Mean Opinion Score (MOS) greater than 3.5) for the different content types over wireless networks. We tested 12 different video clips reflecting different content types. We chose Peak- Signal-to-Noise-Ratio (PSNR) and decodable frame rate (Q) as end-to-end video quality metrics and MPEG4 as the video codec. The work should help optimizing bandwidth allocation for specific content in content delivery networks. Keywords-MPEG4; 802.11b; NS-2; PER; Video quality evaluation I. INTRODUCTION Multimedia services are becoming commonplace across different transmission platforms such as Wi-Max, 802.11 standards, 3G mobile, etc. The current trends in the development and convergence of wireless internet IEEE802.11 applications and mobile systems are seen as the next step in mobile/wireless broadband evolution. Users’ demand of the quality of streaming service is very much content dependent. Streaming video quality is dependent on the intrinsic attribute of the content. For example, users request high video quality for fast moving contents like sports, movies, etc. compared to slow moving like news broadcasts, etc. where to understand the content is of more importance. The future internet architecture will need to support various applications with different QoS (Quality of service) requirements [1]. QoS of multimedia communication is affected both by the network level and application level parameters [2]. In the application level QoS is driven by factors such as resolution, frame rate, colour, video codec type, audio codec type, etc. The network level introduces impairments such as delay, cumulative inter-frame jitter, burstiness, latency, packet loss, etc. Recent work has focused on the wireless network (IEEE 802.11) performance of multimedia applications [3,4,5]. In [6,7,8] the authors have looked at the impact of transmission errors and packet loss on video quality. In [9] authors have proposed a parametric model for estimating the quality of videophone services that can be used for application and/or network planning and monitoring, but their work is limited to videophone. Similarly, in [10] authors have taken into consideration a combination of content and network adaptation techniques to propose a fuzzy-based video transmission approach. In [11] the authors have proposed content based perceptual quality metrics for different content types, whereas, in [12] video content is divided into several groups using cluster analysis [13]. However, very little work has been done on the impact of different types of content on end-to-end video quality e.g. from slow moving (head and shoulders) to fast moving (sports) for streaming video applications under similar network conditions considering both network level and application level parameters. We have looked at the two main research questions in the network level and application level as: (1) What is the acceptable packet error rate for all content types for streaming MPEG4 video and hence, find the threshold in terms of upper, medium and lower quality boundary at which the users’ perception of quality is acceptable? (2) What is the minimum send bitrate for all content types to meet communication quality for acceptable QoS (PSNR >27 dB) as it translates to a MOS of greater than 3.5 [14]? To address these two questions, we first classified the video contents based on the spatial and temporal feature extraction into similar groups using cluster analysis [13]. We then carried out experiments to investigate the impact of Packet Error Rate (PER) and hence, find the threshold in terms of upper, medium and lower quality boundary at which the users’ perception of quality is acceptable and identified the minimum acceptable Send Bitrate (SBR) for the content types. We chose Peak-Signal-to-Noise-Ratio (PSNR) and decodable frame rate (Q) [8] as end-to-end video quality metrics and MPEG4 as the video codec. In the presence of packet loss video quality becomes highly time- variant [15,16]. One of the significant problems that video streaming face is the unpredictable nature of the internet in terms of the send bitrate, and packet loss. We further