628 IEEE TRANSACTIONS ON BROADCASTING, VOL. 54, NO. 3, SEPTEMBER 2008 Effects of an Encoding Scheme on Perceived Video Quality Transmitted Over Lossy Internet Protocol Networks Ron Shmueli, Ofer Hadar, Member, IEEE, Revital Huber, Masha Maltz, and Merav Huber Abstract—We analyze viewer-perceived quality of a compressed video stream, transmitted over a lossy IP network with a quality of service mechanism. The parameters of the encoding schemes include the transmission bit rate, the compression depth, the frame size and the frame rate. We demonstrate that when jointly consid- ering the impact of the coding bit rate, the packet loss ratio and the video characteristics, we can identify an optimal encoding scheme that maximizes viewer-perceived quality. The video content, the compression and the transmission are represented by a vector which contains parameters. Based on subjective tests, we obtain a set of observation pairs of labeled samples , where is the quality class related to the vector of input parameters . To determine the significance of these results, we use the analysis of variance (ANOVA) statistical method, which identifies those factors that cause differences in the averages in the subjective tests results, and determines the significance of the results. Finally, we introduce a novel method to predict an optimal encoding scheme based on canonical discriminant analysis (CDA) for feature classification. Index Terms—Encoding scheme, MPEG, perceived quality, QoE, subjective tests, video QoS, video quality. I. INTRODUCTION D IGITAL video communication systems over Internet Pro- tocol (IP) networks exhibit diverse types of distortions that are strongly influenced by video characteristics, compres- sion methods and transmission capabilities. Assuring quality of experience (QoE) for IP television is rapidly becoming a top pri- ority for vendors and service providers as the market is evolving dramatically and services are now commercially deployed. QoE models and network quality of service (QoS) are crucial for de- livering commercial video services. Since perceived visual information is affected by the net- work differently than data or voice services [1], measuring and modeling network capabilities in relation to visual information is imperative for introducing video services over IP networks. Manuscript received October 4, 2007; revised May 27, 2008. Published Au- gust 20, 2008 (projected). R. Shmueli is with the Department of Electrical and Computer Engineering, AFEKA-Tel-Aviv Academic College of Engineering, Tel-Aviv 69107, Israel (e-mail: rons@afeka.ac.il). O. Hadar, R. Huber and M. Huber are with the Department of Communication System Engineering, Ben-Gurion University of the Negev, Bee-Sheva 84105, Is- rael (e-mail: hadar@bgu.ac.il; revital.huber@gmail.com; merav.huber@gmail. com). M. Maltz is with the Department of Industrial Engineering and Manage- ment, Ben-Gurion University of the Negev, Bee-Sheva 84105, Israel (e-mail: maltzm@bgu.ac.il). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TBC.2008.2001242 The convergence of video, voice and data into triple play ser- vice bundles, running on the same network, is a key strategy for many service providers seeking to establish competitive advan- tages over the residential consumer market in the forthcoming years. QoS involves the ability of a network element to achieve some level of assurance that its traffic and service requirements can be satisfied [2]. QoS protocols provide the mechanism to differentiate traffic, while policy defines how these protocols are used. Furthermore, service-driven QoS provides enhanced QoS based on the content or application. Video QoE is an example of service-driven QoS system. QoE is a subjective measure of performance in a system, which relies on human opinion, and thus differs from QoS which can be precisely measured. Video QoE is composed of the quality of the network de- livery system, the quality of the encoding/decoding components of the network, and human factors. It is affected by the following factors: A. Compression Impairments A lossy compression, such as MPEG, exploits two types of redundancy: spatio-temporal redundancy and psycho-visual redundancy. Examples of compression-related impairments include edge busyness, error blocks, localized smearing, and blocking/tiling [2], [3]. The characteristics of the source video play a crucial role in determining the amount of compression that is possible, and hence, the severity of the compression artifacts [3]. B. Rate Control Rate control means jointly adjusting some coding parameters in order to achieve a target bit rate for the video sequence trans- mission [4]–[7]. The compression parameter of the rate control in compression standards is usually quantization. But two addi- tional parameters can be used for rate control—spatial resolu- tion and frame rate. To evaluate the influence of rate control on the video stream quality, subjective tests are often useful. C. Transmission Impairments Video transmission over IP networks is considered as a real- time application. In our simulations the video streams are trans- mitted through networks using UDP/IP at a constant bit rate (CBR), since that best suits the QoS enforcement mechanism [8]. The most important parameters of transmission networks are one-way delay (delay), instantaneous packet delay variation 0018-9316/$25.00 © 2008 IEEE Authorized licensed use limited to: BEN GURION UNIVERSITY. Downloaded on December 14, 2009 at 12:22 from IEEE Xplore. Restrictions apply.