A no-reference video quality metric using a Natural Video Statistical Model

The demand for high quality multimedia content is increasing rapidly, which has resulted in service providers employing Quality of Service (QoS) strategies to monitor the quality of delivered content. However, the QoS parameters commonly used do not correlate well with the actual quality perceived by the end-users. Numerous objective video quality assessment (VQA) metrics have been proposed to address this problem. However, most of these metrics rely on the availability of additional information from the original undistorted video to perform adequately, which will increase the bandwidth required. This paper presents a No-Reference (NR) VQA algorithm, which extracts a Natural Video Statistical Model using both spatial and temporal features to model the quality experienced by the end-users without needing additional information from the transmitter. These features are based on the observation that the statistics of natural scenes are regular on pristine content but are significantly altered in the presence of distortion. The proposed method achieves a Spearman Rank Order Correlation Coefficient (SROCC) of 0.8161 with subjective data, which is statistically identical and sometimes superior to existing state-of-the-art full and reduced reference VQA metrics.