Objective video quality metric based on data hiding

In this paper, a new no-reference (NR) objective metric based on data hiding is proposed. The metric has the advantage of being fast and not requiring knowledge of the original video contents. The proposed method uses a spread-spectrum embedding algorithm to embed a mark (binary image) into video frames. A t the receiver, the mark is extracted and a measure of its degradation is used to estimate the quality of the video. We used data gathered from psychophysical experiments to help in the design of the video quality assessment system. We evaluated the visibility and annoyance of the impairments caused by the embedding algorithm and estimated the 'best' mark strength for a particular video. The performance of the proposed metric is estimated by measuring its ability to predict the total squared error (TSE) of the host video and the mean observer score (MOS) obtained from naive subjects in a psychophysical experiment. Experimental results show that the proposed metric had a good performance and a good correlation with the MOS.