होम SPIE Proceedings [SPIE Third International Symposium on Multispectral Image Processing and Pattern...

SPIE Proceedings [SPIE Third International Symposium on Multispectral Image Processing and Pattern Recognition - Beijing, China (Monday 20 October 2003)] Third International Symposium on Multispectral Image Processing and Pattern Recognition - A new multiple description coding method over wireless channel

, , , , ,
यह पुस्तक आपको कितनी अच्छी लगी?
फ़ाइल की गुणवत्ता क्या है?
पुस्तक की गुणवत्ता का मूल्यांकन करने के लिए यह पुस्तक डाउनलोड करें
डाउनलोड की गई फ़ाइलों की गुणवत्ता क्या है?
खंड:
5286
साल:
2003
भाषा:
english
DOI:
10.1117/12.538944
फ़ाइल:
PDF, 232 KB
0 comments
 

To post a review, please sign in or sign up
आप पुस्तक समीक्षा लिख सकते हैं और अपना अनुभव साझा कर सकते हैं. पढ़ूी हुई पुस्तकों के बारे में आपकी राय जानने में अन्य पाठकों को दिलचस्पी होगी. भले ही आपको किताब पसंद हो या न हो, अगर आप इसके बारे में ईमानदारी से और विस्तार से बताएँगे, तो लोग अपने लिए नई रुचिकर पुस्तकें खोज पाएँगे.
A New Multiple Description Coding Method Over Wireless Channel*
Xiao Song Wu Chengke Zhang Fang Yan Yaoping
ISN National Key Lab., XIdIan UnIversity. XIan 710071 China
ABSTRACT
A new rnuitipie description selection/separation method for transmission of video bit stream over wireless channel is
proposed in this paper. The method inserts transition frames according to the relative motion between two neighboring
frames, and then divided the video sequence into two descriptions with independent prediction loops. The experimental
results show that the method can help the decoder more quickly recovered from single loss or burst loss compared with
previous niethnd, andprovides more stable an better quality for the reconstruction of video sequence.
Keywords: multiple description coding (MDC), video bit stream, wireless channel, path diversity

1. INTQUCTQN
Limited band-width and serious fading are two main characteristics of wireless channel, which hamper the realization of
multimedia communication of emerging Third and Fourth Generation (3G and 4G) cellular system. In order to combat
the bit error or data loss resulted by fading of transmission signal, many approaches have been proposed, including the
methods applied in channel coding and transmission aspects" , such as forward error correction (FEC) and ARQ etc, and
those applied in source coding aspect such as layered or scalable coding', e.g. MPEG-4 FGS. However, FEC is efficient
only when the number of error does not surpass the threshold of its correcting ability, otherwise the performance will
decrease drastically, and even worse than that ofno FEC. ARQ can be very valuable in the case of a point-to-point link
which has a reliable back channel and sufficiently short round-trip-delay, otherwise the performance degradation can be
quite significant. Layered or scalable approaches essentially prioritize data and thereby support intelligent discarding the

data (the enhancement data can be lost or discarding while still maintaining usable video), but the vid; eo can b

.

completely lost ifthere is an error in the base layer.
Multiple Description Coding (MDC) a no ther approach emerged recently, it attempts to overcome the problem of data
error or loss by coding a signal into multiple bit streams such that any one bit stream can be used to decode a baseline
signal, and any additional bit streams will improve: the quality of the: reconstructed signal. Recent applications of MDC
ideas to video coding are based on temporal subsequences with resync frames4, predictive MD quantizer5, MD transform
coding6, and multiple: states7 etc.
in the literature7, a system combining MDC with path diversity was proposed. in which the video is coded into multiple
independently decodable streams, each with fts own prediction process and sent over different path. Although the system
can effidently combaterror. propagatinnthe performance of fast moving sequence is not as good as that of slow moving
sequence because it averagely partition the frames to odd and even subsequences for independent prediction ioop at the
encoder. In this paper, we propose a new multiple description selection/separation method for video coding, better
performance will be achieved for both slow and fast video sequence.
This paper is organized as follows. Section 2 presents the system architecture of the proposed method and describes the
new multiple description. selection/separation method in. detail. in. section.3 , the performance of the proposed scheme is
examined conipa red' to otheralgnrithm. Th conclusion &fthe paperwill be present d in Section 4.

2. PROPOSED SYSTEM ARCHITECTURE
2.1 System overview
The proposed system architecture is shown in figure 1 . The raw video stream taken from camera or source storage is
partitioned into two subsequences of frames, which are coded into two separate bit streams by multiple description
selection/separation. Each stream forms a description with different prediction loop and is independently decodable from
the other. The encoder consists of two separate conventional encoders, or an encoder which stores the last two previously
*

334

This project is s'upported by the National Natural Science Foundation of China (No. 60002007).

Third International Symposium on Multispectral Image Processing and Pattern Recognition,
Hanqing Lu, Tianxu Zhang, Editors, Proceedings of SPIE Vol. 5286 (2003)
© 2003 SPIE · 0277-786X/03/$15.00

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

coded frames and chooses which previous coded frames to use to form the prediction for the current frame. Then
different streams are transmitted over different channels undergoing independent error effects to minimize the chance
that both streams are lost. Similarly, the decoder also consists oftwo separate decoders or a single decoder that alternates
which previous decoded frame it uses to perform the prediction. If there are no errors and both transmission stream are
received correctly, then both streams are decoded to frames which are interleaved for final display. If a stream has an

error, there will be error propagation for that stream since all following P frames before the next I frame will be
erroneous due to prediction with wrong reference frame.
MUitil)1-' l-'S(Ii1)tiOll co(liIlg
However, the other independently decodable stream can

still be accurately and straightforwardly decoded to

-

I

rate-. There are- no- other distortions b-y using this ap-prdach,

_________

which is different from conventional methods which are
forced to either freeze the video or attempt to estimate the
unknown video by performing some form of concealment;

storage

either

___

rcamera

produce usable video at the cost of the reduction of frame

/____ Encodei

Raw video sirewia

H
H
_____j

_$___]

Channel]

DisplayL1 l('IllF

of which can lead to significant distortion,

especially if there are many P frames before the next I
frame.

R

Decoder

! p....- J—1Decoder
t

1

Ma] t

i Ii] P ileiii ii iou ulecodiuug

2.2 Multiple description selection/separation
Previous method7 averagely divided original video stream

Figure 1: System. architecture of the proposed- method
into two subsequences with even frames and odd frames to
form two descriptions, which is efficient when there exists only slow moving of object or background. While for fast

moving sequence, the muftiple description. separation will further increase the magnitude of the motion vector of
macroblock between two neighboring frame, which may cause the encoder to frequently select intra mode to encode the
macroblock, thus lead to drastic fluctuate of coding rate. Furthermore, if an error occurs in large motion vector, the
reconstructed video quality will decrease drastically, and it will be very difficult for the decoder to recover form the error.
As a resalt, a new- multiple description selection/separation method fo c transmission of video bit stream over wireless

channel is proposed in this paper. The method inserts transition frames according to the relative motion between two
neighboring frames, and then divided the video sequence into two descriptions. The larger the relative moving is, the
more transition frames are inserted in one video sequence. The method .increases the relativity between frames by
inserting transition frame, which can not only improve the reconstructed quality of video with only one correctly
received description due to bad channel conditions at the receiver, but also overcome the problem of drastic fluctuate of
coding rate when there exists fast moving in original video sequence, thus may decrease the realization complexity of
post processing application such as rate control. The detailed method for multiple description separation is described as
follows.
Firstly, a criterion is determined for frame insert.
Define SAD as the parameter to describe the relative motion degree between two neighboring frames, which is

- X(x,y,t -

1616

SAD(x,y) =

In order to take into account both global motion and local motion, the macroblock is used as unit to the calculation of
SAD. So SAD, denotes the relative motion ofthe ith macroblock between two frames. X(x,y,t) and X(x,y,t —1) denote
the pixel value ofpoint (x, y) in current and previous frame respectively.
Given a threshold ':S-AD of SAD, there are two criterions that can reflect the relativity between two neighboring frames.

(2)

They are max SAD1 > TsI)
!E(1Nu1j )

arid

NMB

> Tf

(3-)

Where NMB (M . N)I(l 6 . 1 6) , which denotes the number of macroblock contained in one frame with M in vertical
resolution and N in horizontal resolution. N -, denotes the number of macroblocks whose value of SAD are larger
AD, >1 cw

than a threshold of T SAD .

denotes another threshold of fraction.

The equation 2 reflects the local motion of some macroblock while the equation 3 reflects the global motion of large part
o-f macro-bloc-ks. Transition frame- wi-Il- be- inserted when either of two- equatidns is satisfied.

Proc. of SPIE Vol. 5286

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

335

Secondly, the algorithm of calculating the inserting frame is determined.
Considering the complex ofthe algorithm, there are three approaches for inserting the frame.
I.
inserting the average value oftwo neighboring frames
2. inserting a copy ofprevious frame
3 . inserting a copy of next frame.
Since the multiple description separation is finished before the video coding process of motion estimation, there is not
any information about the motion of macroblock in the whole video sequence. At the same time, the motions between
frames are often very complicated. (e.g. Scene change, single object moving, multiple objects moving etc.) Therefore,

inserting the frame by directly averaging two neighboring frames may not be suitable. Sometimes, it iiay introduce
additional error to video encoding process. Consequently, the second and the third approaches may be used according to
some criterions in our method.

Assuming F. denotes the macroblock motion fraction of the jth frame relative to the (j- l)th frame ,

FMVj— N

SAL >Ts,•jj

and

I NMB . Then the algorithm can be described as follows:

If FM > FMV(/+l) then
inserting a copy ofthe (j+1) th frame;
else
inserting a copy ofthe -1) th frame.

End if

3. SIMULATION RESULTS
The effectiveness of the proposed multiple description video coding system is examined for football(352 X 240
pixel/f30f/s) sequences over wireless channel. Two independently stream with their own prediction process are produced

after multiple description separation. Two paths are selected for path diversity, such that each system is sent over a
different path. The tests assume the existence of an ideal path diversity system, which provides two paths with
independent losses, and each loss leads to the corruption of the entire frame. In order to help the decoder recover from
errors, the system should estimate the lost frame by some approaches, including low-complexity approaches such as
simply replacing the lost frame by a correctly decoded frame, or a MC correctly decoded frame, or a more sophisticated
MC interpolation (MC-I) algorithm, e.g. compute the motion field across a subset of correctly decodable past and future
frames from corrupted and uncorrupted streams, and apply appropriate linear or nonlinear filtering along the motion
trajectories. The recovery should also account for covered and uncovered areas within the frame by appropriately
choosing to use only future or previous frames to estimate the appropriate areas. In order to compared with previous
methods, the proposed system estimated the lost frame by computing the motion between the neighboring correctly
received (previous and future) frames in other description and performing a motion compensated interpolation, in
contrast to conventional single description system, which estimates the lost frame as the last correctly decoded frame.
Two different types of losses were examined: one is single loss, which cortesponds to the loss of a single entire frame;

the other is burst loss of
1 OOms

duration,

corresponds

which

Single Loss(Football)

to the loss of

threefram:s.For simplicity
single
taken

sequence is received correctly.

performances under
different types of loss of the
the
system,
proposed
conventional
single
description
(SD) system with
.
.
The

single path and previous
multiple description (MD)
a

coding system7 are shown in

.

°
2

first

sequence . while. the second

336

32

loss or burst loss is

place in the

Burst Loss(Football)
34

z

::

.

:

•

g 24
:

20
18

•
is

.

.

.

...

z

24

45

55

.

:•

18

.

es

Frame Number

Figure 2: reconstructed video quality
of single loss

single description
:: •' ••/ • conventional
.. previous multiple description

Th
14

35

.:

22
20

.

—•— conventional single descnption
I —— previous multiple description
proposed multiple description
25

1\(\J

. 26

:

26

22

..

.

28

J

proposed multiple description

0

10

20

30

45

50

Frame Number

Figure 3: reconstructed video quality
of burst loss

Proc. of SPIE Vol. 5286

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

figure2 and figure3 respectively.
With the proposed system
.

.

15

a .single. frame loss leads
to
.

system beyond that of a single loss, and it leads to

are

illustrated in figure 4 when all of the methods are operating at

—k--— previous multiple description

proposed multiple description

:
.

' A..
•1 \•

. 12

•

••I\

/\

\i

..

°
15

additional about 1 .6dB loss for previous MD system, and 3dB for
conventional SD system. The compression ratios of three systems
the same PSNR. It is shown that both the previous MD method and

I

.— conventional single description

14

system the loss is about 6dB and for the conventional SD system it's
almost 14dB. Since the conventional SD system has a higher

proposed

—----

•

approximately a 2dB loss in quality, while for the previous MD

percentage of intra-coded macroblocks, it converges slightly faster
to the point of complete recovery (assuming no loss during this
period). However it is more vulnerable to losses. The lOOms burst
loss (loss of three consecutive frames) has minimal affect on the

—

*

oo

I"

!a
°

'

.

is

.

45

55

60

Frame number

Figure 4 : Compression ratio ofthree methods

our proposed method have a lower compression ratio than
conventional SD method because of the separation process of multiple description which further increasing the bits of P
frames. Although the transition frames inserting in our method add additional redundancy to the multiple description
coding, they decreases the relative motion between two neighboring frames thus only have slightly drop in compression
ratio compared to previous method.

4. CONCLUSION
A new multiple description selection/separation method for transmission of video bit stream over wireless channel is
proposed in this paper. The method inserts transition frames according to the relative motion between two neighboring
frames, and then divided the video sequence into two descriptions with independent prediction ioops, which may further
improve the reconstructed quality of video with only one correctly received description due to bad channel conditions at
the receiver compared to previous method. The characteristics ofthe proposed system are described as follows:
1 . The process of multiple description separation is very simple, so it's easy for realization.
2. No back channel is needed compared with other methods such as NewPred in MPEG-48 and Reference Picture
Selection (RPS) in H.263 version 2, in which reliable back channel is necessary.
3 . The method is applicable to many types of video sequences including slow and fast moving sequences at the cost of a
slightly increasing in processing complexity.
4. Although the method adds additional redundancy to the multiple description coding for its inserting transition frames,
the relative motion between two neighboring frames is decreased, thus the compression ratio only has a slightly drop
compared to previous method.
The experimental results show that the method can help the decoder more quickly recovered from single loss or burst
loss compared with previous method, and provides more stable and better quality for the reconstruction of video
sequence.

REFERENCES
I.

2.
3.
4.
5.
6.
7.
8.

W.Tan and A.Zakhor, " Error control for video multicast using hierarchical FEC," IEEE Inter. Conflinage Processing, pp.40 1-

405. Oct. 1999.
Y.Wang and Q.Zhu, "Error control and concealment for video communications: A review," Proceedings qf the IEEE, pp.974997, May 1998.
Hayder M. Radha, Mihaela van der Schaar and Yingwei Chen, "The MPEG-4 fine-grained scalable video coding method for
multimedia streaming over IP", IEEE trans. on multimedia,3(1): pp.53—68, Mar. 2001.
S.Wenger, G.Knorr, J.Ott, and F.Kossentini, "Error resilience supports in H.263+," IEEE Trcins. On Circuits and Svstenis for
Video Technology, pp.867-877, Nov. 1998.
V.Vaishampayan and S.John, "Interfrarne balanced-multiple-description video compression ICIP. Oct.1999.
A.Reibrnan, H.Jafarkhani, Y.Wang, M.Orchard, and R.Puri, "Multiple description coding for video using motion compensated
prediction," IEEE Inter. Confln2age Processing, Oct. 1999.
J.Apostolopoulos, "Reliable video communication over iossy packet networks using multiple state encoding and path diversity,"
Proc. SPIE Conf Visual Comm. Image Process, pp.392-409. Jan. 2001.
Coding ofAudio-Visual Objects: Visual (MPEG-4 Video), Committee Draft ISO/IEC 14496-2, Mar. 1999.

Proc. of SPIE Vol. 5286

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

337