Joint source channel distortion modeling for MPEG-4 coded video sequences

We developed a joint source-channel distortion model for predicting distortion in Moving Picture Expert Group (MPEG) 4 coded video sequences due to quantization and channel errors. Important components of video coding such as differential coding, variable length coding, run-length coding, and motion estimation and compensation are taken into account in this model. In addition, different error resilience tools in MPEG-4 standard such as data partitioning and packetization (resynchronization) are also used in the modeling process. Distortion expressions are derived for different types of data in the I and the P coded frames. The amount of distortion propagated to future frames due to errors at any point in the bitstream is also predicted by the expressions in this model. Model parameters are derived using a training database of video sequences. These parameters are then used in conjunction with the local statistics of the data in the video packets to determine the amount of distortion in different test video sequences. Test simulations are carried out and the model results are compared to the test simulation results. Results show that this model predicts the amount of distortion introduced in the MPEG-4 coded video sequences due to quantization and channel errors to within 1.5 dB PSNR of that obtained via test simulations.