Survey of Dirac: A Wavelet Based Video Codec for Multiparty Video Conferencing and Broadcasting - Intelligent Video Event Analysis and Understanding - page 219

terms of the perceptually-
the difference between the

-weighted error fourth-power error E(q,4), resulting from
e original and the quantised coefficients.

m

5.2 Wavelet Transfo orm

After the motion com
compensation are treated
have one luminance and
dimensional arrays of data
component data as follow
The data arrays are w
divided into subbands.
Then they are quantize
Finally the quantized d
The architecture of coeffic

mpensation is performed, the residuals of motion
almost identically to Intra picture data. In both cases, w
d two chrominance components in the form of two
a values. There are three stages of picture (frame or field
ws:
wavelet-transformed using separable wavelet filters an

n-
we
o-
d)

nd

ed using RDO quantizers in the reference encoder.
data is entropy coded.
cient coding is shown here:

Fig. 5 Coefficient coding arc chitecture [8]

As can be seen, each w
and the entropy coding
coded bands. This does l
parent-child relationships
The only difference betw
residual coefficient codin
subband of Intra picture c
At the decoder side, th
entropy coding is decode
reconstructed to produce
inverse transform produc
encoder has to maintain a
be used as reference fram
and the decoder will not r
The discrete wavelet tr
in numerous references. I

wavelet subband is coded in turn. Both the quantisatio
of each band can depend on the coding of previousl
limit parallelisation, but the dependences are limited t
, so some parallelisation/multi-threading is still possibl
ween Intra picture coefficient coding and Inter pictur
ng lies in the use of prediction within the DC wavel
components.
he three stages of the coding process are reversed. Th
ed to produce the quantised coefficients, which are the
the real values. Then, after undoing any prediction, th
ces the decoded picture component. The reference Dira
local decoder within it, so that compressed pictures mu
mes for subsequent motion compensation else the encode
remain in sync.
ransform is now extremely well-known and is describe
In Dirac it plays the same role of the DCT in MPEG-2 i

on
ly
to
le.
re
let

he
en
he
ac
ust
er

ed
in