Video Quality Estimation for Mobile Video Streaming

For the provisioning of video streaming services it is essential to provide a required level of customer satisfaction, given by the perceived video stream quality. It is therefore important to choose the compression parameters as well as the network settings so that they maximize the end-user quality. Due to video compression improvements of the newest video coding standard H.264/AVC, video streaming for low bit and frame rates is possible while preserving its perceptual quality. This is especially suitable for video applications in 3G wireless networks. Mobile video streaming is characterized by low resolutions and low bitrates. The commonly used resolutions are Quarter Common Intermediate Format (QCIF,176×144 pixels) for cell phones, Common Intermediate Format (CIF, 352×288 pixels) and Standard Interchange Format (SIF or QVGA, 320×240 pixels) for data-cards and palmtops (PDA). The mandatory codec for Universal Mobile Telecommunications System (UMTS) streaming applications is H.263 but the 3GPP release 6 already supports a baseline profile of the new H.264/AVC codec. The appropriate encoder settings for UMTS streaming services differ for various streaming content and streaming application settings (resolution, frame and bit rate). In the last years, several objective metrics for perceptual video quality estimation were proposed. The proposed metrics can be divided into two main groups: human vision model based video metrics and metrics based on empirical modeling. The complexity of these methods is quite high and they are mostly based on spatial features, although temporal features better reflect the perceptual quality especially for low-rate videos. Most of these metrics were designed for broadband broadcasting video services and do not consider mobile video streaming scenarios. The goal of the presented research is to estimate video quality of mobile video streaming at the user-level (perceptual quality of service) for a large set of possible codec settings in 3G network and for a wide range of video content. Measures were derived that do not need the original (non-compressed) sequence for the estimation of quality, because such reference-free measures reduce complexity and at the same time broaden the possibilities of the quality prediction deployment. New reference-free approaches are presented for quality estimation based on motion characteristics. Moreover, this thesis provides a detailed comparison of recently proposed models for video quality estimation.