No-reference video quality evaluation for high-definition video

Abstract

A no-reference video quality metric for High-Definition video is introduced. This metric evaluates a set of simple features such as blocking or blurring, and combines those features into one parameter representing visual quality. While only comparably few base feature measurements are used, additional parameters are gained by evaluating changes for these measurements over time and using additional temporal pooling methods. To take into account the different characteristics of different video sequences, the gained quality value is corrected using a low quality version of the received video. The metric is verified using data from accurate subjective tests, and special care was taken to separate data used for calibration and verification. The proposed no-reference quality metric delivers a prediction accuracy of 0.86 when compared to subjective tests, and significantly outperforms PSNR as a quality predictor.

No-reference video quality metric for HDTV based on H.264/AVC bitstream features

“…The results show that the proposed metric outperforms our previous no-reference metric in [1] slightly with respect to the Pearson correlation and the RMSE, but especially well with respect to the Spearman rank order correlation. This is not surprising, as we build our model only for H.264/AVC, compared to [1], where we also considered alternative, wavelet-based coding technologies and therefore the metric is not optimized only for H.264/AVC. Also, we achieve better results than the metric in [5].…”

“…Besides the Pearson and Spearman rank order correlation coefficents, we also provide the root mean squared error (RMSE) between predicted and actual visual quality. For comparison, we included the results of our no-reference metric presented in [1], but also the results of two well-known full-reference video quality metrics: SSIM [11] and the VQM according to Annex D of ITU-T J.144 [12]. For the latter, the general model was used.…”

“…If we used the same data for training and validation, it would lead to overly optimistic prediction models as discussed in [1]. Therefore we perform a cross validation and split the available data set into four different subsets and apply the PLSR on each of these subsets.…”

“…Thus ratings do not reach the boundaries of the scale, but saturate much earlier. Therefore, we correct the prediction valuesŷ slightly, using a fixed sigmoid nonlinear correction in order to emulate this behavior [1]. The sigmoid correction ofŷ is given aŝ…”

“…With appropriate noreference algorithms, features that describe certain visual properties of the video sequences, like blocking or bluring, are extracted and then combined in a no-reference metric. One example of such a metric, especially for HDTV, is our previous contribution in [1].…”