Low-Complexity No-Reference PSNR Estimation for H.264/AVC Encoded Video

Abstract

We present a no-reference (NR) PSNR estimation method which is based on only two bitstream features (average bitrate and mean quantization parameter of the I-frames). The low computational complexity of the proposed method makes it suitable for in-network real-time applications. The NR metric achieves a Pearson correlation of 0.99 for individual videos and a RMSE of approximately 1 dB PSNR on average. We additionally investigate the effect of various encoding configurations on the PSNR and show the robustness of our method towards these. Finally, we incorporate the proposed metric into an example application and demonstrate that only a minor performance loss is observed compared to the reference scheme which assumes the availability of true PSNR information.