Estimating PSNR Without Reference for Real H.264/AVC Sequence Intra Frames

The paper describes a novel metric for video quality measurement. The metric is capable of estimating peak signal-to-noise ratio (PSNR) values of pictures in compressed bit streams conforming to the H.264/AVC standard. It is designed and has been tested for intra predicted pictures with varying encoder settings. The metric uses an artificial neural network to estimate PSNR examining the compressed bit stream only. In other words, we do not need to decode the actual pixels within the pictures being evaluated. We present a concept of the metric and its performance results in this paper.