First-order distortion estimation for efficient video streaming at moderate to high packet loss rates

This paper reports on advances in end-to-end distortion estimation for streaming of pre-compressed video. It builds on the “first-order distortion estimate” FODE) which employs a Taylor expansion of the distortion about the operating point of zero packet loss rate (PLR), while accounting for packet loss, error propagation and error concealment at the decoder. Improved estimation can be achieved by similar expansion about operating points of non-zero PLR, but would normally involve impractical complexity to evaluate the impact of many possible packet loss patterns. Instead, we propose a new method where FODE is combined with the “recursive optimal per-pixel estimate” (ROPE) to provide accurate, low complexity estimates via expansion about an arbitrary reference PLR vector. The reference vector need not be uniform and can exhibit the expected profile of effective PLR versus packet position in the group of packets (GOP). The method further accounts for the commonly neglected error propagation across GOP boundaries due to error concealment at the decoder. The main experimental results show significant gains in estimation accuracy at medium to high packet loss rates. Some preliminary results on embedding the estimate within a simple delivery policy selection system are also provided.