Objective quality prediction model for lost frames in 3D video over TS

This paper proposes an objective model to predict the quality of lost frames in 3D video streams. The model is based only on header information from three different packet-layer levels: Network Abstraction Layer (NAL), Packetised Elementary Streams (PES) and Transport Stream (TS). Transmission errors leading to undecodable TS packets are assumed to result in frame loss. The proposed method estimates the size of the lost frames, which is used as a model parameter to predict their objective quality measured as the Structural Similarity Index Metric (SSIM). The results show that SSIM of missing stereoscopic frames in 3D coded video can be predicted with Root Mean Square Error (RMSE) accuracy of about 0.1 and Pearson correlation coefficient of 0.8, taking the SSIM of uncorrupted frames as reference. It is concluded that the proposed model is capable of estimating the SSIM quite accurately using only the lost frames estimated sizes.