PARAMETRIC PACKET-LAYER VIDEO QUALITY MODEL FOR IPTV M.N. Garcia, A. Raake Quality and Usability Lab Deutsche Telekom Laboratories, Berlin University of Technology Berlin, Germany ABSTRACT This paper presents a parametric packet-layer model for pre- dicting perceived quality of High Definition (HD, 1920x1080) and Standard Definition (SD, 720x576) videos for IPTV ser- vices. The model can be applied both for network planning and service monitoring. It takes as input the video resolution, packet-layer information such as bit-rate, packet-loss-rate and burstiness factor, and information on the encoder/decoder set- tings such as the number of slices per frame and the packet- loss-concealment. Addressed degradations are compression artifacts related to H.264 and MPEG2 encoding, and trans- mission errors leading to slicing errors or freezing, depending on the applied packet-loss-concealment. Extensive subjec- tive video quality tests have been conducted for measuring the perceived quality of the original and degraded video se- quences. The model is developed based on the obtained test results. The model prediction shows a correlation of 0.98 for HD and 0.96 for SD with the mean subjective ratings. 1. INTRODUCTION In order to achieve a high degree of user satisfaction for video services like IPTV, perceived video quality needs to be pre- dicted both in the network planning phase and as part of the service monitoring. Quality assessment can be achieved us- ing perception tests or by instrumental methods that yield es- timates of video quality as perceived by the user. Since per- ception tests are time consuming, costly, and do not allow assessing the quality during real-time service operation, in- strumental assessment methods are often preferred. Those in- strumental methods are based on video quality models. A classification of instrumental method types and related stan- dardization activities are given in [1], [2] and [3]. In this paper we present a parametric packet-layer model. In case of network planning, the signal is not available and in- stead of measured packet-layer information planning assump- tions are made. In case of service monitoring, the parametric model does not require access to the original signal and has low computational complexity since it does not require de- coding or partial-decoding of the bitstream. It is thus appro- priate for real-time service monitoring. It is also suitable for encrypted data since it requires only packet-header informa- tion as input. The model is part of the framework presented in [3]. It extends the video quality model described in [3] to bursty losses for both freezing and slicing. The resulting models, one for HD and one for SD, are thus valid for vari- ous codec (H.264 and MPEG2), at various constant bit-rates, for uniform and bursty losses and for freezing 1 and slicing 2 . In the case of slicing, we applied two encoding settings: one in which each video frame is split into 3 slices (”3sl”) and one in which there is 1 slice per Macro-block line (”sl”), thus many slices per frame. The two different numbers of slices per frame yield two different types of perceptual effects. Several studies are relevant to be mentioned in this con- text. In particular, Verscheure et al. proposes in [4] a video quality model based on the average bit-rate and packet-loss- rate, and Yamagashi et al. developed in [5] a packet-layer model taking as inputs the average bit-rate and packet-loss- frequency. The approach in [5] is close to the one presented here and will be compared to our models in a future study. Models found in the literature are generally not fully applica- ble to our scope since they either address other applications and/or smaller formats (ITU-T Recommendation G.1070, [6]), or uniform loss only [4], or H.264 only and other types of error- and packet-loss-concealment [5]. Several further studies have been conducted on the qual- ity impact of bursty losses. In [7] and [8], which address resp. QCIF and CIF formats, isolated losses are found to have bet- ter perceptual video quality than bursty losses. In [9], which addresses SD format, bursty losses are found to have gener- ally better quality than isolated multiple losses. Those at first sight contradicting results show that the distribution of losses has a strong quality impact, but, following a deeper analysis of those results, that this impact depends on factors such as the video format, the packet-loss-concealment, the number of slices per frame, the Group-Of-Picture (GOP) structure and 1 in case of packet loss, the picture freezes until the next intact I-frame arrives; the frames in between are skipped in the considered IPTV case. 2 a slice typically corresponds to a certain area of the image that – if af- fected by loss – the decoder fills with data from the same, previous or follow- ing video frame. The decoder can resynchronize based on the next available slice header.