Packet-header based no-reference quality metrics for H.264/AVC video transmission

In video streaming, bitstreams are often transmitted in best-effort IP networks where impairments such as congestion and varying delay often cause artifacts in the decoded video. In this scenario, packet losses should be detected as early as possible in the transmission chain, preferably inside the network, where perceptual video quality metrics are needed to assess the users Quality of Experience (QoE). These metrics should have high efficiency and low complexity, especially to satisfy the demanding real-time and scalability requirements (many video bitstreams) of the network nodes. This paper proposes low complexity no-reference video quality metrics for H.264/AVC video transmissions in packet-based networks. These video quality metrics predict subjective scores that represent well the decoded quality with features extracted from the headers that encapsulate compressed video data. The evaluation shows that packet-header features allow an accurate quality prediction when packet losses occur in video transmission.