Scalable H.264 Wireless Video Transmission over MIMO-OFDM Channels - Advanced Concepts for Intelligent Vision Systems - page 41

6 Priority Mapping of Scalable Layers

We considered two different mappings of the temporal and FGS layers of SVC
into the scalable layers μ i . Let us assume that the number of temporal layers
is T . For a GOP size of 8, as used here, we have T = 4. For the first mapping,
which we call the Temporal-SNR mapping, the first L

1layers( μ 1 ,...,μ L− 1 )
are the base layers (FGS0) of the frames associated with the lowest to the highest
temporal level in decreasing order of importance for video reconstruction. So,
L

−

1= T and the number of scalable layers is L = T +1. The FGS layer of all the
frames in a GOP are defined as a single layer μ L of even lesser importance. The
Temporal-SNR distortion estimation method uses exactly the same priorities as
the Temporal-SNR mapping, so we used it for our experimental results for this
mapping. The second mapping is the SNR-temporal mapping. For this mapping,
there are two layers, base and enhancement for each of the T temporal layers. In
this case, the FGS layer of the lower temporal layer has more importance than
the base layer of the higher temporal levels. Thus, there are a total of L =2 T
scalable layers. For the SNR-Temporal mapping, we used the SNR-Temporal
distortion estimation method, as it uses exactly the same priorities.

−

Fm

46.5

SNR−TEMP
TEMP−SNR

46

45.5

45

44.5

44

43.5

43

42.5

42

41.5

8 9 10 11 12 13 14 15 16 17 18

SNR

Fig. 2. Performance of the cross-layer optimization using the Temporal-SNR and the
SNR-Temporal mappings of scalable layers (“Foreman” sequence)

7 Experimental Results

For experimental results, the “Foreman” and 'Akiyo” video sequences are en-
coded at 30 fps, GOP=8 and constant Intra-update (I) at every 32 frames. We