From the statistical results of Table 6.2 and the following sample frames, it is obvious that all error control techniques perform well for the given packet loss conditions. The effects of packet loss are shown in Figures 6.6b, 6.7b, and 6.8b. These images feature severe distortions including block artifacts and other spatial deformations, caused by the absence or incorrect decoding of either macroblock information, or motion vectors information.
As discussed in previous chapters, the temporal prediction error concealment performs well only on low-motion sequences, and only for individual errors (not error bursts). The artifacts that are caused because of the use of the previous frame's information, are frustrating for moving parts of the image. This is clearly seen in Figure 6.6c for the helmet and part of the face, in Figure 6.7 for the mother's hand, and in Figure 6.8 for the subject's eyes. However, these sequences are played at 15 fps, and these errors are hardy notifiable. The problem with error repair, is that it is not actually repair, as the error will be preserved for next frames if the encoder is not notified, or intra-coded blocks do not appear. This is reflected on its relatively low average performance statistics of Table 6.2.
On the other hand, forced intra-coding provides a much more robust bitstream. The more intra-resyncs inserted, the better picture quality is obtained, as shown in Figure 6.5. However, this comes at the price of significantly increased bandwidth requirements. It should be stressed that the frames illustrated in Figures 6.6d, 6.7d, and 6.8d, are not intra-coded.