Information Technology Reference
In-Depth Information
encoding the binary value on their
p
-th bit-plane. Finally, in the normalization
pass, the significance of the remaining previously non-significant coefficients is
coded. Again, when a coefficient becomes significant with respect to
T
, its sign
is also coded. The application of the coding passes is illustrated in Figure 6.
The symbols signaling the significance, the signs and the refinements bits for
different coefficients are coded using a combination of run-length coding and con-
text-based binary adaptive arithmetic coding. For further details, the reader is re-
ferred to [20].
Since the wavelet coefficients are coded in a bit-plane by bit-plane manner, a
quantized version of the original wavelet coefficients in each block
C
can be ob-
tained by simply truncating the resulting compressed bit-stream
B
. This property
is exploited in Tier 2 of the codec. In Tier 2, the code-block bit-streams
B
are
cut-off at a length
l
, whereby
l
is determined such that the requested bit-rate,
resolution and region of interest are met and the decoder-side distortion is mini-
mized. This is achieved using Lagrangian rate-distortion optimization [20, 21].
5 Experiments
In this section, the H.264/AVC Intra and Motion JPEG 2000 coding standards are
experimentally evaluated. A first series of experiments reports the compression
performance for the two coding systems applied on 1080i25 and 720p50 video
material. The employed test material consists of 5 video sequences from the SVT
test set [22] in 1080i25 and 720p50 formats: "CrowdRun", "ParkJoy", "Duck-
sTakeOff", "IntoTree" and "OldTownCross". All sequences were converted from
the original SGI format to planar Y'C
b
C
r
4:2:2 video with 10 bits per component
using the commonly employed conversion tools of [23].
For H.264/AVC Intra the JM reference software version 12.2 was used [24].
Although this is not the latest version available, this option was taken due to erro-
neous operation of more recent versions on 4:2:2, 10 bit video material. For Mo-
tion JPEG 2000 the Kakadu software version 6.1 [25] was used. The employed
parameter configurations for both codecs are summarized in Table 1 and Table 2
for 720p50 and 1080i25 material respectively.
With respect to Motion JPEG 2000, it must be observed that the standard was
not particularly designed to target interlaced material. In the codec's normal con-
figuration, all interlaced frames are coded in the same manner as a progressive
frame, which corresponds to the frame coding mode in H.264/AVC. However,
field coding can easily be supported by separating the fields of each interlaced
frame and coding each field as a separate tile (this corresponds to H.264/AVC's
field coding mode).
The results obtained with Motion JPEG 2000 on both frame and field coding
modes are reported. The compression performance of the two coding systems is
evaluated by measuring the Peak Signal to Noise Ratio (PSNR) at the following
bit-rates: 30, 50, 75, 100 Mbit/s. The average PSNR per frame is calculated as
