An adaptive network control scheme for region-based hybrid coding algorithm

L SCHEME

FOR REG1

CODING

ALGORITHM*

Hsu-Tung Chen, Liang-Gee Chen, Sheng-Chieh Huang, Tsung-Han Tsai, Hao-Chieh Chang

Department of Electrical Engineering, National Taiwan University,

No.

1,

Roosevelt Road, Sec. 4,

Taipei, Taiwan, R.O.C.

ABSTRACT

Based on the packet loss rate and the frame loss rate experiment in the practical Intemet environment, a novel adaptive network control scheme is proposed for region-based hybrid coding

algorithm. In comparison with the sourcdchannel coding [ 11, this

adaptive network control scheme is presented to reduce 50%

frame loss rate during the Intemet weak day use condition and

58% frame loss rate during the Internet peak day use condition.

In the video conference system, the compressed multimedia data

is transmitted to receiver by network. The network need to

guarantee the real time transmission, low delay, and low packet

loss rate. In the past researches, there are the sourcdchannel

coding [l] and automatic repeat request (ARQ) [2] mechanism to

solve the packet loss problem. In this paper, a novel adaptive network control scheme is applied to the region-based hybrid coding algorithm [3] for solving the packet loss problem of the multimedia compressed data.

ASED HYBRID CODING

ALGORI~HM

There are two important properties in the video conference system. In first, camera is fixed in video conference application. Secondly, a partocipant appears in front of an unchanged background, and moves almost slowly. The information of the background can be removed for reducing the computation loading and the network congestion.

'This work was supported by National Science Council under

Grant NSC 87-2215-E-002-028.

0-7803-4455-3/98/$10.00

0

1998

IEEE

The following example explains the region-based hybrid coding algorithm. A circle object moves in consecutive frames from lower left to upper right as shown in Fig.

1

(a) and Fig 1 (b). The difference of two frames is shown in Fig. 1 (c). It is obvious that the changed region is the two circle objects. It is only necessary to encode the two objects.

The region-based hybrid coding algorithm is shown in Fig. 2. In

the coding algorithm, the I frame is Intra-Frame(n) and the P

frame is Frame(n)

-

Predict-Frame(n-I) as shown in Fig. 3. The

I frame coding algorithm and P frame coding algorithm are

described in the following:

2.1

I frame Coding Algorithm

The I frame is encoded as still image. The following is the I

frame coding algorithm:

Step

I:

Iframe is divided into many 8x8 blocks.

Step 2: For each block in Iframe, do the following processing:

{ do-SxS-DCC quantization; run length coding; VLC;

add to transmission buffer container; )

2.2 P frame Coding Algorithm

Step I: Pframe is divided into many 8x8 blocks.

Step 2: Calculate MAE (Mean Absolute Error) for per block. Step 3: For each block in Pframe, do the following processing:

{ $(MAE > threshold)

{ do-8xS-DCZ

quantization; run length coding;

VLC;

add to transmission buffer container;))

The MAE (Mean Absolute Error) and threshold are used to

identify the changed blocks in

P

frame.

3. ADAPTIVE NETWORK CONTROL

SCHEME

The design of the adaptive network control scheme is based on

the packet loss rate of the UDP protocol experiment and the frame loss simulation of the region-based coding algorithm on

Internet. The two experiments are described in the following:

3.1 The Loss Rate

of

the

UDP

Protocol

Experiment:

This experiment is performed at peak use of network and heavy load of cicsun06 (load=1.2) as shown in Fig.4. The TD is time division (ms) between the two successive packet transmissions. There are two results shown in this figure. In first, the larger packet size will have higher probability of packet loss at the same time division. Secondly, the longer time division will have lower packet loss rate at the same packet size.

3.2 The Frame

Loss Simulation of the Region-

based Coding Algorithm

Without retransmission of the lost frame, the Fig. 5 represents that the packet arrival process is highly correlated with the successive packet drops. When there are more than three successive frames loss, the PSNR value of the frame will decline quickly and video quality will be poor.

3.3 Adaptive Network Control Scheme Using First

Order Markov Model

According the above two experiments result, a novel adaptive network control scheme is proposed for the region-based hybrid coding algorithm. In this scheme, the successive packet loss

condition can be detected and the frame loss rate can controlled with the adjustment of the frame transmission threshold.

The packet

loss

condition can be detected with the transmission status between the transmitter and the receiver. After receiver gets a video packet, it sends a grant to transmitter immediately.

So, the packet loss condition can be counted and the first order Markov model is used to simulate the behavior of the network as shown in Fig. 6 . In this model, there are two states and four parameters described as following:

PS:

the state of a successful packet.

PL: the state of a loss packet.

P S S the probability of a successful packet following a successful packet.

P S L the probability of a successful packet following a loss packet.

PLS: the probability of a loss packet following a successful packet.

PLL: the probability of a loss packet following a loss packet. The two equations can be got as following:

PSS

+

PSL = 1

P L L + P L s = l

Based on the above model, a novel network control scheme is designed to adapt the video transmission. This scheme degrades

or upgrades the server transmission rate depending on feedback message from receiver on the control channel.

When the PSL and the PLL are small, it is hard to become the PL

status from the

PS

status, but easy to become the

PS

status from the PL status. The two conditions ( PSL

>

cpsl ) or ( PLL > cplf )

are used as criteria. The cpsl and cpll criteria parameters are the

probability of the packet loss and used to detect the satus of the packet loss. The cpsl and cpll criteria parameters are adjusted

according the necessity of the users.

If the packet loss rare is high, the rhreshold of the region-based hybrid coding algorithm is increased to reduce the data rate of the

changed blocks in the P frame. When network is in weak use, the threshold is decreased to get high video quality. The adaptive network control scheme is described as following:

if (( PSL > cpsl ) or ( PLL

>

cpll)) ( if (threshold

<

800) threshold

+=

50; else ( if (framerate

>

1) framerate = frame - I ; ) ) else ( if (threshold

>

100) threshold -= 50; else

(if (framerate < IO)

framerate = frame

+

1 ;) )

The threshold is bounded between 150 and 800. If the threshold is equal to 800 and packet loss rate is still high, theframerate will be reduced to low down the packet loss rate. The transmitter adjusts PSS and PLL according to statistical data from receiver. Based on the practical Intemet experiment, the network condition can be represented by sampling data each 10 seconds.

4. EXPERIMENTAL RESULTS

The adaptive network control scheme is tested on the 120 seconds salesman sequence. The spatial resolution is 320x240 for 8-bits gray level. The Internet transmission experiments are carried out between the video5 workstation and the cicsun06 workstation as shown in Table. 1. and in Fig. 7.

As shown in Fig. 8.. the different statistical time is used to

understand network condition by applying the adaptive control scheme. When the 10 seconds are chosen as the statistical time, the packet loss rate is the lowest.

Table. 2. shows the comparison of frame loss rate for various scheme. Testing time is at 08:OOam during periods of weak day use. Table. 3. represents the frame loss rate during network peak day use. Because the network control scheme can adjust coding strategy according network condition, the frame loss rate is lower

than sourcdchannel coding in [l]. In comparison with the sourcdchannel coding in [ 11, this adaptive network control scheme is presented to reduce 50% frame loss rate during the Internet weak day use condition and 58% frame loss rate during the Internet peak day use condition.

5.

CONCLUSIONS

A novel adaptive network control scheme for region-based hybrid

coding algorithm is proposed in the above sections. The first order Markov model is used to control the network behavior. It is to meet the bandwidth of the network by minimizing data rates of transmitter. From the results of the real Internet experiments, this adaptive network control scheme is superior to the source/ channel coding.

REFERENCE

[l]. J. C. Bolot et al., "Adaptive Error Control for Packet Video in the Internet," Proceeding of IEEE Intemal Conference on Image Processing, 1996.

[2]. Y. Chen etal., "A robust Coding Scheme for Packet Video," IEEE trans. on Commun., Vol. 40, No. 9, pp. 149 1

-

1501, 1992.

[3]. H. T. Chen, P. C. Wu, Y. K. Lai, and L. G. Chen, "A multimedia Video Conference System: Using Region-based Hybrid Coding," IEEE trans. on Consumer Electronics, Vol. 42, No. 3, pp. 781-786, 1996,

Figure 1. Ermple of region b;w hybrid coding.

.

?I

Figure 6. The swtc diagnm of p r k e t transmission in the nelwork.

*

(

Fipre 7. Testins environment

Table I. Host computers used in our experiment.

Mechanism Frame loss rete

Figure 2. Block dkegnm of region b i w hybrid codins system.

Proposed No network control SourcdChannel Cdinp. l l I

5% 6%

3%

I P P P I P P

Figure 3, Fnme type

Fnmc Ius> retc

I IJo.126.16.2 !muter

icsun06.cic.edu.tw ]I4Olo.l26.2S.6 Itransmitter

16.67% 11%

1.9%

L

L. Y Y. .u .D u *.

hub.

Figure 4. Packet loss rate for vuriour packet stze

Figure 5. Frame loss simulation result of Region-bilx hybrid coding

(Threshold=MX)). There are two successive frame IOU ul f m c number 22 and 23 Thcrc arc thrcc successive fnme loss at fnme number 32.33 Md 34. There arc

four succes?itve ireme loss x f r a m number 42,4394. and 45. There arc live

successive f m e loss ill frame number 52. SI. 54.55. and 56.

nu

r---~

Figure n. Statistical T i m d x c ) VI Loss Ratio (cpsl = 0.1: cpll = 0.1).