> Replace This Line with Your Paper Identification Number (double-click Here to Edit) <

The paper contributes to video quality assessment of delivered content without reference. We propose a no-reference video quality assessment metric taking into account the behavior of the human visual system. The proposed metric called Weighted Macro-Block Error Rate (WMBER) is based on macro-block error detection and is weighted by visual saliency maps. Both measures are extracted from a partially decoded H.264 AVC stream. First of all, we propose a new saliency map fusion method to improve the spatiotemporal saliency model. Then a supervised learning method called Similarity Weighted Average is considered to predict subjective MOS from objective video quality metric. The Similarity Weighted Average method is improved in order to be adapted to a training database or a content. The performance of the proposed metric is evaluated on two subjective experimental databases from LaBRI and IRCCyN. The results are compared with two Full-Reference metrics MSE and SSIM. The evaluation shows that the proposed metric provides an accurate prediction of subjective measures.