Blind estimation of the QP parameter in H.264/AVC decoded video
Marco Tagliasacchi, Stefano Tubaro · Virtual Community of Pathological Anatomy (University of Castilla La Mancha) · 2010
No-reference video quality monitoring algorithms rely on data collected at the receiver side. Typically, these methods assume the availability of the bitstream, so that motion vectors, coding modes and prediction residuals can be readily extracted. In this paper we show that, even without the availability of the bitstream, the decoded video sequence can be reverse engineered in order to reveal part of its coding history. Specifically, we illustrate a method for blindly estimating the quantization parameter (QP) in H.264/AVC decoded video on a frame-by-frame basis. We demonstrate by means of extensive simulations the robustness of the proposed algorithm. We discuss its usefulness in the field of video quality assessment (e.g. to perform blind PSNR estimation) and we provide an outlook on video forensics tools enabled by the proposed method (e.g. to detect temporal cropping/merging).