Cryptography Reference
In-Depth Information
0
1
0
2
2
5
5
1
3
3
4
4
6
6
Fig. 3.17.
Spatial sub-bands after a typical 2-level wavelet decomposition [22, 23].
uses a scaled down version of the motion vectors, which are inadequate from
time to time. One solution to this problem is the 2D+t structure, which is
also called in-band MCTF [20]. We will discuss the 2D+t(+2D) structure in
Section 3.3.4.
3.3.2 Spatial and SNR Scalability
To achieve spatial scalability, a separable 2-D wavelet transform is applied to
both the temporal low-pass and the high-pass frames. Similarly to JPEG2000,
when only the low-resolution images are to be reconstructed, we decode the
corresponding lower sub-band coe
cients. A typical 2-level 2-D wavelet de-
composition is shown in Fig. 3.17.
To provide the SNR scalability with fine granularity, an embedded zero-
tree coding [6] can be used after the spatial decomposition. In addition to [6],
other methods such as embedded zero-block coding (EZBC) [5] and embed-
ded block coding with optimized truncation (EBCOT) [7] are also commonly
used for coding the wavelet transform coe
cients. This is done to reduce the
temporal redundancy among successive frames. These coding techniques are
used in the VidWav software and can be further extended to a 3-D structure
such as 3-D EBCOT as proposed in [21].
3.3.3 Motion Scalability
The compressed bit-stream includes both the texture and the motion informa-
tion. The bit-stream for the texture part can be almost arbitrarily truncated.
However, the motion information can not be easily partitioned in the original
wavelet video proposal, Motion Compensated Embedded Zero Block Coding
(MC-EZBC) [5]. This is because the motion information can be large and can
amount to consume a large percentage of the transmitted data in the low
bit rate applications. This leads to few bits available for transmitting texture
data. This results in poor subjective image quality. To solve this problem,
motion information should also be represented in a scalable manner.
In [28], a scalable representation of motion information is proposed for the
MC-EZBC scheme. Furthermore, in [29], the motion information is partitioned
