Efficient Reduced-Reference Video Quality Meter

We present an efficient reduced-reference video quality assessment method based on local harmonic strength (LHS) feature. LHS is based on harmonics gain and loss information derived through the discriminative analysis of harmonics strength computed from gradient pictures. Harmonics gain and loss correspond with the two most prominent compression artifacts, namely blockiness and blurriness. To make the method effective for continuous monitoring of video quality, it is necessary to minimize the reduced-reference overhead data whilst keeping the accuracy of the quality meter sufficiently high. For the LHS-based method, we recommend quantization of the LHS feature, temporal sampling, and block selection method, either with or without segmentation-based block classification. A typical reduced-reference low overhead data rate (around 160-400 bps) with good prediction performance (0.86-0.88 Pearson correlations against mean opinion score) for broadcast type video sequences in VQEG Test Phase-I data sets has been achieved by the proposed method.