The choice of input video sequences determines the range of the potential applications for the scheme. The attributes that are important here, are the picture size and frame rate of the sequence.
Picture sizes that are acceptable for entertainment video, are at least the size
of current PAL and NTSC formats. These are approximately the size of 4CIF, which
has both horizontal and vertical dimensions double the size of CIF, see
page ![[*]](cross_ref_motif.gif)
for details. For personal communications applications,
picture sizes of CIF and even QCIF provide adequate resolution for head and
shoulders images, and fit better to small screen mobile terminals.
Frame rates in films and television are in the range of 24-30 fps. If lower values are used, the eye can distinguish the succession of images, which causes a frustrating flickering effect. However, for smaller resolutions, and for sequences with low motion and not major scene changes due to camera pan and zoom, smaller frame rates can be used without notice from the user.
The selection of the above attributes has to be judicious for two main reasons. First, the bandwidth requirements of high quality video could become unbearable for the network infrastructure. Second, the real-time decoding and display of TV-quality pictures in current desktop or mobile terminals is not a trivial task yet, and has a direct impact on the decoder's cost.
Taking into account these facts, and as this work will evaluate mainly schemes for reliable personal communications, the input sequences that have been selected are all head and shoulders QCIF images, sampled at 15 fps. Three sequences with different motion and spatial detail characteristics were selected in the conducted experiments, described in the next chapter.