Asimetrik kodlama ve aktarım optimizasyonu

It is well known that the human visual system can perceive high frequencies in 3D, even if that information is present in only one of the views. Then, the best 3D stereo quality may be achieved by asymmetric rate allocation to the reference (left) and auxiliary (ri1ght) views. However, the question of what should be the level of asymmetry between reference and auxiliary views without disturbing the perceived quality and whether the rate reduction for the auxiliary view should be achieved by spatial resolution reduction (coding a downsampled version of the video followed by upsampling after decoding) or quality (QP) reduction is an open issue. Subjective tests indicate that when the reference view is fixed at sufficiently high quality (i.e. 38 dB), the auxiliary view can be encoded around 31 dB without losing much on the perceived stereo video quality depending on the type of utilized 3D display technology. The threshold is ~31 dB for a parallax barrier display and ~33 dB for a polarized projection display. Beyond this threshold, symmetric coding starts to perform better than asymmetric coding in terms of perceived stereoscopic video quality. In addition to those, the result of subjective tests shows that users prefer SNR scalability over spatial scalability on both parallax barrier and polarized projection displays above the threshold..Although it is more desirable to have more than 1 MGS NAL unit(NALU) since number of extraction points increase as the number of MGS NAL units increases, as number of NALUs increases, the coding efficiency decreases. We have tested the R-D performance of streams including only 1 NALU with streams including 2 NALUs. We used our own Bitstream Extractor program to extract the streams at various bitrates and compare their R-D performances.Additionally we present the performance of different multi description generation algorithms. The streams of each descriptions are generated by MD generator program that is coded by us. Moreover we coded a channel simulator program to investigate the R-D performance of these descriptions under different network conditions (channel loss probabilities).