This paper proposes and evaluates a no-reference video quality assessment metric that considers both the effects of lossy H.264 video encoding and packet losses over IP networks. The algorithm produces quality scores computed from elements extracted from the bitstream
(DCT coefficients and their corresponding quantizing steps) and information about the packet losses positioning, derived from the arriving packet headers. Three different packet loss models are analyzed along the paper: 1) a simple model that accounts for the
packet loss ratio; 2) a model that considers the frame type where the correspondent losses occur; 3) a model that considers both the frame type and a set of measures taken from the packet loss pattern. The latter showed the best performance and its quality predictions
are well correlated with the quality scores resulting from subjective assessment.