No-reference PSNR estimation algorithm for H.264 encoded video sequences

This paper proposes a no-reference PSNR estimation algorithm for video sequences subject to H.264/AVC encoding. The proposed method explores statistical properties of the transformed coefficients, which can be modeled by a Cauchy or Laplace probability density function. The distribution's parameters are computed from quantized coefficient data received at the decoder, combining maximum-likelihood with linear prediction estimates. Since the proposed algorithm has no knowledge about the original sequences, it can be used as a no-reference metric for evaluating the quality of the encoded video sequences. When compared with recent state-of-the-art algorithms proposed for the same purpose, it has shown better PSNR estimation accuracy in a set of video sequences subject to different encoding rates.