Cryptography Reference
In-Depth Information
motion compensated filtering to achieve the temporal scalability. Other possi-
ble structures will be discussed in Sec. 3.3.4. To achieve the spatial scalability,
the 2-D wavelet decomposition is done for each frame after the MCTF. To
further reduce the bit rate and generate a scalable bit-stream, the wavelet
coe
cients are coded using a scalable entropy coder such as zero-tree coder
or other types of arithmetic coder. By means of proper extraction techniques
a single bit-stream with scalable spatial, temporal and SNR parameters can
be produced. Similar wavelet video coding schemes but with specific features
have been suggested by several other researchers [20, 21].
Historically, the wavelet-based schemes fell short in video quality in the
MPEG standard competition in 2004. An ad hoc group in MPEG with a
coined name VidWav was formed and has continued to study the wavelet
video coding technique for the past two years. Based on the software originally
developed by Microsoft Research Asia, this group produced a common sim-
ulation platform for testing and improving wavelet video algorithms [22, 23].
Additional reports on the recent VidWav group activities are summarized in
[24] and [25].
In the following subsections, we give a brief overview of the basic con-
cepts of the wavelet video scheme. The MPEG VidWav evaluation software
will often be used for illustration. Although there are a number of on-going
research works outside MPEG, our description focuses only on the MPEG
related wavelet video projects. An interesting phenomenon is that over the
past 3 years, several tools such as MCTF originally proposed for the wavelet
schemes are now included in the AVC-based SVC with an open-loop concept.
These tools have been discussed in the previous sections and we will minimize
the overlap here. Some AVC tools such as the AVC base-layer are tested under
the wavelet structure and the Variable Block-Size Motion Compensation.
3.3.1 Temporal Scalability
To provide the temporal scalability, the wavelet-based scheme also adopts the
MCTF as described in Section 3.2.1. The MCTF concept was first proposed
for wavelet coding [5]. Typically, Haar or (5, 3) wavelets are used.
To improve the accuracy of the motion field so as to improve coding per-
formance, several techniques for better motion estimation have been proposed
in [26, 27]. Particularly, in [27], a novel structure, known as Barbell-Lifting,
is presented. The basic idea is to use a barbell function for generating the
prediction/update values in the lifting structure. Specifically, for each pixel in
the high-pass frame, the prediction value is obtained by using a set of pixels
as the input to the barbell function. It has been shown that the prediction
using the barbell function offers a superior performance when compared to
the conventional scheme using a single-pixel. It often reduces the mismatch of
motion in the prediction and update steps. However, in the conventional t+2D
structure, the motion vectors are estimated at the highest spatial resolution.
In reconstructing the spatial low-resolution images the temporal prediction
