Complexity Scalable Hybrid End-to-End Distortion Estimation for Conversational Video Streaming

In conversational video streaming applications, periodic I-frame coding is not allowed due to strict low delay requirement. For this scenario, we propose a novel hybrid paradigm, which takes advantage of both accurate pixel-level estimation and efficient block-level estimation. Specifically, error propagated (EP) distortion from the last N coded frames are exactly calculated per pixel for high accuracy, while EP effect beyond the frame limit N are estimated per macroblock for low complexity. By varying the frame limit, flexible tradeoffs between complexity and performance can be achieved, rendering our scheme especially suitable for wireless video streaming, where stringent power resource is always a challenge. Simulation results justify the effectiveness of the proposed hybrid ED estimation scheme.