Cryptography Reference
In-Depth Information
Layer (in QCIF resolution) is coded at a frame rate of 15 Hz with H.264/AVC
using the structure of hierarchical B pictures. The spatial enhancement layer
in CIF resolution is coded at a frame rate of 30 Hz using the MCTF with
4 decomposition stages. Each spatial layer has one SNR base layer and two
FGS layers. For each SNR layer, the left hand side of Fig. 3.16 specifies its bit
rate at maximum frame rate.
As shown, the QCIF layer has a maximum bit rate of 120kbits/s. By
truncating the FGS layers, one can obtain a quality scalable bit-stream with
a bit rate ranging from 40kbits/s which implies a minimum guaranteed quality
to 120kbits/s. Also, a bit-stream of reduced frame rate can be extracted by
dropping the B pictures in a dyadic manner. For instance, one can obtain a
QCIF sequence of 7.5 Hz by dropping the frames labeled as B
2
. A frame rate
of 3.75 Hz can be obtained by discarding all the B pictures. Together with the
SNR scalability, we can have a QCIF bit-stream with bit rate ranging from
10kbits/s to 120kbits/s and associated frame rate starting from 3.75 Hz to
15 Hz.
For the combined scalability, the same techniques can be applied in order
to obtain a CIF bit-stream with a bit rate ranging from 20kbits/s to 480kbits/s
and corresponding frame rate starting from 3.75 Hz to 30 Hz. Due to the inter-
layer prediction, the decoding of the CIF layer requires the bit-streams of the
QCIF layer. As a result, the bit rate of the CIF layer at 30 Hz starts from
160kbits/s, which includes the bit rates of the QCIF bit-streams. Particularly,
in the example of Fig. 3.16, the 2nd FGS layer of the QCIF resolution is only
used for the refinement of the QCIF layer. It is not exploited for the inter-layer
prediction. Such a bit-stream is referred to as a Dead-Substream. It has been
proved that using all the FGS layers for the inter-layer prediction may not be
optimal for coding e
ciency [17, 18]. Dead-Substream provides the flexibility
to select an optimal truncation point for the inter-layer prediction.
3.3 Scalable Approach Using Inter-Frame Wavelet
By contrast to the aforementioned AVC-based scheme that basically employs
the hybrid coding structure, a wavelet-based scheme using (t+2D) structure
was proposed in [5, 19]. Similar to the AVC-based approach, the wavelet-
based scheme can produce a fully embedded bit-stream that simultaneously
supports spatial, temporal, and SNR scalability. However, a major difference
between the AVC-based approach and the Wavelet-Based approach is that
all the predictions in the latter case are conducted in an open-loop manner.
The open-loop prediction provides more flexibility for bit-stream extraction
and is more robust with respect to transmission errors. A recent variation can
include a closed-loop prediction structure but it is not yet popular.
Fig. 3.2(b) shows the basic structure of a wavelet-based scheme using the
t+2D structure. Within each GOP, MCTF is used for temporal decomposi-
tion. Particularly, the temporal low-pass frames are recursively generated via
