Compressed-domain-based Transmission Distortion Modeling for Precoded H.264/AVC Video Fan li Guizhong Liu IEEE transactions on circuits and systems for video technology, 2009

Outline  Introduction  Transmission Distortion Modeling  Experimental result and discussion  Accuracy and Complexity analysis  Example  Conclusion

Introduction  When the video sender drops packets due to congestion, or when packets are lost in the channel, transmission error occurs and would further propagate to its subsequent frames along the motion prediction path.  Traditional methods, such as ROPE, are pixel domain based distortion estimation, which are computationally inefficient.

Transmission Distortion Modeling  Since the decoding resynchronization is done at the slice header for the H.264 video, the loss of any packet in one slice will cause unsuccessful decoding of the whole slice. Thus the transmission distortion caused by the transmission errors can be calculated as follows:  From the expected loss probability of the slice  Assumed a simple error concealment strategy Temporal Replacement (TR) by copying the information of the entire slice at the corresponding location of the latest decoded frame.

Transmission Distortion Modeling  Estimation of D L (f,n)  (f, n, i) and (f, n, i) be the ith reconstructed pixel of the nth slice in the fth frame at the encoder and decoder f-1f f A B

Transmission Distortion Modeling  Estimation of RFD(f, f -1, n)  Q i,j represents the relative motion intensity of the jth block in the ith inter-coded MB

Transmission Distortion Modeling  Estimation of D R (f,n)  MB is intra-coded

Transmission Distortion Modeling  MB is inter-coded

Experimental result and discussion  Selection of W i The relative value of the distortion is estimated by the CDB model. Therefore, we only focused on the proportion between Wi of the intercoded prediction to that of the intra-coded prediction for both the 16*16 and 4*4 modes. Compare W i = (β, 1.1β, 1.2β), (β, 1.2β, 1.4β), (β, 1.3β, 1.5β), and (β, 1.4β, 1.6β)

Experimental result and discussion V2(W i = (β, 1.2β, 1.4β)) fits best with the actual RFD and smallest deviation

Experimental result and discussion  Estimation of transmission distortion at PER=10%  Estimation of time-varying channel (wireless network)

Experimental result and discussion  influence of the bit rate to the accuracy of the CDB model  Average Error Rate

Experimental result and discussion

 Complexity Analysis  Feature extraction  Experiment results show that the processing time of the CDB model is only 42.8% of that in the ROPE and LPP approaches.  Distortion estimation  CDB model is based on the MB level estimation. On the contrary, the ROPE and LPP approaches are based on the pixel level estimation, and computation is operated per pixel. Number of operations in the CDB model is approximately 1.34%-1.75% of the number in the ROPE and LPP approaches.

Example  A base station delivers the video streams to three mobile users and uses TDMA based scheduling…

Example  Decision function  Object function

Example  CLD : using the LPP model as the decision function  CDBRA scheme outperforms the CLD scheme by 1.46 dB, and the no optimization scheme by 2.82 dB.

Conclusion  A compressed domain approach to the transmission distortion modeling has been proposed. The approach has a much lower computational complexity when compared with that in the conventional pixel-domain- based methods and also provide fine accuracy and robustness.