Available online at www.sciencedirect.com Available online at www.sciencedirect.

com

Procedia Engineering
www.elsevier.com/locate/procedia

ProcediaProcedia Engineering 15 000000 Engineering 00 (2011) (2011) 3421 3427

Advanced in Control Engineering and Information Science

An Energy Efficient and Load Balancing Distributed Image Compression Algorithm in WMSNs
Feng Tiana,Jia Liub,Enyan Sunc,Chuanyun Wangd a*
School of Computer Science Shenyang Aerospace University Shenyang, China

Abstract According to the problem that resource-constrained always restricts image transmission in the wireless multimedia sensor networks (WMSNs), this paper combines with theory of parallel computing and puts forward a kind of energy efficient and load balancing distributed image compression algorithm based on the JPEG2000. The algorithm introduces node competitive factor and communication cost function for the purpose of making the most of the limited resource to transmit more images and maintaining load balancing in the image compression process. In addition, the algorithm encodes the information separately which is divided accord to the frequency after wavelet transform, in order to relieve the load of the cluster head and prolong the network life time. Further simulations demonstrate that the algorithm can ensure a certain quality of the reconstructed image, reduce 30% energy consumption of the cluster head node, greatly balance the load of network nodes and effectively prolong 15% of the network life time.

2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of [CEIS 2011]
Keywords:WMSNs; Image compression; Distributed; Load balancing; Low power consumption; JPEG2000

1. Introduction The image compression in Wireless multimedia sensor networks (WMSNs) has its own characteristics, node must meet the requirement of real-time compression in the network, and compression ratio of single node is extraordinary high. Traditional image compression algorithms are computation complexity and high overhead, they are not suitable for the resource-constrained WMSNs. Huaming Wu etc proposed distributed image compression which is based on the discrete wavelet transform, and gave the two kinds of data exchange programs in the realization programs, respectively as shown in Fig. 1 (a) and Fig. 1 (b). There are three levels of decomposition. s is the source node, d is the destination node and ci is the cluster head of cluster i. In method 1, the decomposition of each level

*Foundation item: Natural Science Foundation of Liaoning Province (20082011); Shenyang Science and Technology Program (1091185-1-00). Author: Feng Tian (1958 - ), Male, Shenyang, Professor, Ph.D, Doctoral Tutor, E-mail: tianfeng5861@163.com.

1877-7058 2011 Published by Elsevier Ltd. doi:10.1016/j.proeng.2011.08.641

3422 2

Feng et al/ al. / Procedia Engineering 15 (2011) 3421 Feng TianTian etProcedia Engineering 00 (2011) 000000 3427

wavelet includes the row transform and the column transform, and all nodes which participate in each level of wavelet must do round-trip wireless transceiver twice. Because the image information contains a large amount of data, twice round-trip wireless transceiver substantially increase the energy consumption of a single node and decrease the network lifetime. Method 2 solves wireless transceiver frequently by the block compression, but it does not have clear clustering and corresponding scheme which can keep the network load balancing. In response to these problems, we propose an algorithm to achieve JPEG2000 image coding energy efficient and load balancing, which is suitable for WMSNs. To achieve this, the algorithm improves method 2 data exchange way, encodes high frequency information (LH, HL and HH) separately according to sub-band, and introduces node competition factor and communication cost function for the algorithm, and the new network topology model shown in Fig. 1 (c). The simulation results show that this algorithm can reduce the energy consumption of the cluster head node, make the network load more balance and prolong the network lifetime meanwhile ensures the quality of image.

Fig. 1. (a) Method 1;

(b) Method 2;

(c) Our improved Method

2. Energy consumption model In the study for parallel wavelets, the researchers usually assume that the energy consumption of data passed between the processor closes to zero or very small, but this assumption can not be established in the WMSNs. According to the literature [1], nodes in the process of a wireless transceiver transmit k bits of data per amount of energy is
Eelec * k + k * fs * d 2 , d < d0 ETX (k, d ) = ETX elec (k ) + ETX amp (k, d ) == 4 Eelec * k + k * mp * d , d d0

(1)

and the energy consumed in reception per bit is ERX (k ) = ERX elec (k ) = Eelec * k In addition, the node energy consumption of the data integration and processing is Edata . 3. Algorithm Description

(2)

This paper divides the wireless sensor nodes into camera nodes and ordinary nodes, and assumes that: (1) Every node in the network has a unique ID number; (2) All camera nodes and all ordinary nodes do not change their location after deployment; (3) All nodes can perceive their residual energy; (4) Every Node can calculate the distance between sender and itself according to the received signal strength RSSI; (5) Sending node can adjust their own send power according to the distance of receiving node.

Feng Tian etet al/ /Procedia Engineering 15 (2011) 3421 3427 Feng Tian al. Procedia Engineering 00 (2011) 000000

3423 3

3.1. Clusters build The image compression algorithm which was proposed by Huaming Wu uses LEACH protocol which periodically chooses the cluster head by equal probability. LEACH protocol has some shortcomings, and is not suitable for the WMSNs. This paper introduces the distance factor and the load factor in the selection of cluster head. The distance factor adjusts distance between the camera node and cluster head nodes, as well as the distance between sink node and cluster head nodes, in order to achieve the purpose that completed the distributed image compression in relatively small energy transmission consumption. Through the load factor, we can predict the load capacity of the node and strive for that the low energy consumption in every round and more energy residual of the node has the greater probability to become cluster head node. The camera nodes and cluster head nodes use single hop communication mode. We add physical weights into distance factors to control the distance between them. The physical location weights is d d (i, s ) w1 = max (3) d max d min Among them, the camera node is s, d max and d min is the maximum and minimum distance of node in the network from the camera, and d (i, s) is the distance between any node i with the camera. In order to adjust the position of the cluster head node in image processing tasks with sink node and reduce energy of the image data transmission consumption after compression, this paper adds logical location weight into the distance factor and defined as l w2 = 1 i (4) lmax li is the hops of cluster head node to the sink node, lmax is the maximum. In order to balance influence of two weights, we add the normalization coefficient k ( k = 0.3 ), and defines the distance factor is W (i ) = (1 k ) * w1 + k * w2 , (5) According to the defect that LEACH doesnt consider energy factor, this paper focuses to consider the residual energy of nodes and the average energy consumption in each round when chooses the cluster head. We suppose initial energy of node i is E0 (i ) in round r, the current energy is Ecurrent (i ) . If Ecurrent (i ) Eave ( r ) , the node will has the opportunity to compete for cluster head node. The average energy consumption of the node which can compete for cluster head node in the round r is defined as 1 n Eave ( r ) = [ E0 (i ) Ecurrent (i )] (6) N iG' Based on Eave _ r , we estimate load capacity of node in the next round. The node load factor is calculated as
Ecurrent (i ) On the basis of above formula, the probability of the node elected as the cluster node is
p [ *W (i) + E forecast (i)], i G ' T (i) = 1 p[r mod(1/ p)] 0, i G '

E forecast (i ) = 1

Eave _ r

(7)

(8)

Among them, p is percentage (the number of cluster head nodes in all nodes), r mod(1/ p) is number of nodes elected as cluster head in current round, G ' is nodes set which satisfies competition requirement and is not elected as cluster head node in this round. In addition, energy factor is primary consideration in

3424 4

Feng et al/ al. / Procedia Engineering 15 (2011) 3421 Feng TianTian etProcedia Engineering 00 (2011) 000000 3427

the choice of cluster head, the paper sets (0 < 1) in the formula (10), values of is determined according to specific application. This paper also considers the defects of the protocol used in the original image compression scheme and introduced communication cost function when clustering, the communication cost function is
cos t (ci , i ) = f1 (d (ci , i )) + f 2 (d (cinext , i ))
f1 (d (ci , i)) = d (ci , i ) , d i max f 2 (d (cnext , i ) = d (cinext , i) di max

(9) (10)

Where ci is cluster head node, cinext is its next hop node, d (ci , i ) and d (cinext , i ) is the distance between i and cluster head, d i max is the the maximum distance between i and cluster head. For any node select cluster head which has the smallest cos t (ci , i ) to join in. 3.2. Image compression In view of the actual situation of resource constraints in WMSNs and characteristic of JPEG2000 coding algorithm is computation complexity and high storage requirement. This paper improves centralized encoding algorithm, a large task of image compression is divided into several sub tasks and the nodes which accomplish the same task run the same sub-algorithm. Compared to the original distributed algorithm, the new algorithm encodes information separated by sub-band after the each level of wavelet transform, shown in Fig. 2, in order to achieve the purpose of reducing energy of the cluster head and maintaining load balancing. Finally, we bring together the data of all the nodes participated in image compression task and get the solution of initial task.

Fig. 2 Wavelet-based image compression process

This paper divide the whole network into two layers, all the camera nodes as the first layer, the common nodes participated in image compression task as the second layer and randomly distributed in a certain area. Image compression and transmission process is as follows: (1) The camera node notifies the base station when captures image, the base station broadcasts this information in its effective communication radius, the nodes meet competitive requirement according to the formula (5) and (7) compute distance factors and load factors, then elect cluster head node on the basis of T(i);

Feng Tian etet al/ /Procedia Engineering 15 (2011) 3421 3427 Feng Tian al. Procedia Engineering 00 (2011) 000000

3425 5

(2) Cluster head nodes based on the distance between the camera, self-organization to form a virtual cluster head chain, and the rest nodes select the cluster head according to the communication cost function (9) and form a cluster; (3) Camera node pretreatment image and divide it into M blocks, select M nodes with the largest residual energy in the cluster, then image blocks achieve the first level wavelet transform separately; (4) M nodes participated in wavelet transform transfer processed data to the next-hop cluster head; (5) For each tile, the next hop cluster chooses a part of the results (LL) and sends it to a group of processing nodes in its cluster, the rest results (LH, HL and HH) is assigned to three cluster nodes, independently encoded and sent to the next cluster head; (6) Under such circumstances, the final compressed image reaches the sink node when the image compression ratio achieved. In addition, we believe that for a sensor network, the hop between source node and sink node is large enough to single cluster and one level wavelet decomposition, so the node closed to the sink node is unlikely to have much computational load. 4. Simulation In this section, we perform four sets of simulation experiment that compare three algorithms (centralized algorithm, original distributed algorithm and our distributed algorithm) on four performance metrics: total energy consumption, image quality, network lifetime and node load. We employed the communications energy model of literature [1] and considered that all nodes were placed randomly in an area of 500 500 m and the initial energy of all nodes is 10J. The camera nodes arranged in central location of the rectangular region and the base station's initial position is (0, 0). First of all, in order to obtain the optimal value of parameter in our distributed algorithm, we analyze the effect of parameter on total energy consumption in three conditions. The level of wavelet transform is 5 and the value of bpp is 0.1. We can observe in the Fig. 3 (a), the total energy consumption is the most minimum when the value of parameter is close to 0.2, so we set = 0.2 in follow experiments.
11

100

our distributed algorithm (n = 200)

10.9

our distributed algorithm (n = 500) our distributed algorithm (n = 1000)

The num ber of im age successful transm itted

90

T otal netw ork energy consum ption (J)

10.8

80

10.7

70

centralized algorithm (L = 1) original distributed algorithm (L = 1) our distributed algorithm (L = 1) centralized algorithm (L = 5) original distributed algorithm (L = 5) our distributed algorithm (L = 5)

10.6

60

10.5

50

10.4

40

10.3

30

10.2

20

0.1

0.2

0.3

0.4

Parameters u

0.5

0.6

0.7

0.8

0.9

10

100

200

300

400

500

600

700

800

900

1000

Number of nodes

Fig. 3 (a)

effect on total energy consumption of network (b) The change of Network life time followed with the number of node

In this paper, successful collection and transmission the number of 512 512 gray image is assessment measure of the network life time. The curve (Fig. 3 (b)) shows the network life time changes with the number of node in the case of three conditions when the level of wavelet transform is 1 or 5.

3426 6

Feng et al/ al. / Procedia Engineering 15 (2011) 3421 Feng TianTian etProcedia Engineering 00 (2011) 000000 3427

It is observed that network life time largely constraints by energy consumption of the camera node in the case of centralized algorithm, affected by the changes of node density is not obvious. Successful acquisition and transmission the number of image only related with the energy of camera node. When we tried the original distributed algorithm, the network life time was increased obviously along with the number of node increased. Compared to the original algorithm, in the case of that node deployed densely, our algorithm can transmit more than 10 to 15 images, and more effectively prolong the network life time. In order to transmit image effectively and low energy consumption in resource-constrained WMSNs, this paper adopts block compression, but this way inevitably lead to degradation of image quality. Reconstructed image under the three algorithms was shown in Fig. 4. It is observed that the quality loss of tiling with sixty-four tiles under our algorithm (Fig. 4 (b)) is evident in the case of low bit rate compared to the centralized compression (Fig. 4 (a)). We do the same experiment for the original algorithm (Fig. 4 (c)) reconstructed image appears the same block effect. When the number of tiles is small or when the bit rate of compressed image is not very low, we do the experiment under our algorithm again (Fig. 4 (d)). The experimental result shows that distributed algorithm is still applicable in WMSNs image compression if the number of tiles is small or not very low bit rate, and improvement of post processing technology can eliminate block effect.

Fig. 4 Compression of Lena image

We also evaluate load of cluster head nodes and ordinary nodes. The number of collected and transmission image is 10 in per round and the level of wavelet is 5 in the experiment. We consider two conditions that respectively deployed 100 and 500 nodes, and do the same experiment with the original algorithm and our algorithm. The result (Figure. 5 (a)) shows that our algorithm transfer part load of the cluster head to the cluster member nodes more effectively, compared to original algorithm, the decrease of energy consumption is significant, and the network load more balanced. Anything Else, we randomly choose 10 nodes in the case of deployment 500 nodes when we assess the load of ordinary node. The result is shown in Fig. 5 (b). After 50 images are transmitted successfully, compared to original algorithm, our algorithm can make residual energy tend to balance with the network running.

Feng Tian etet al/ /Procedia Engineering 15 (2011) 3421 3427 Feng Tian al. Procedia Engineering 00 (2011) 000000
10
5

3427 7
4

original distributed algorithm(n=100) our distributed algorithm(n=100)

9 3.5 3.5

The energy consum ption of cluster head nodes per round(J)

original distributed algorithm(n=500) our distributed algorithm(n=500)

4.5

8 3 7 3

Node residual energy (J)

Node residual energy (J)

2.5

Node residual energy (J)

1 2 3 4 5 6 7 8 9 10

2.5

3.5

1.5

1.5

3 1 2 1

0.5 1
2.5

0.5

1.5

2.5

3.5

4.5

The number of rounds executed

10

10

Node number (Initial)

Node number (Original algorithm)

Node number (Our algorithm)

Fig. 5 (a) The total energy consumption of cluster head node

(b) The change of node energy in image compression process

5. Conclusion and future work Efficient compression and transmission of images in resource-constrained WMSNs is considered. This paper puts forward a kind of energy efficient and load balancing distributed image compression algorithm based on the JPEG2000. Theoretical analysis and simulation results show that the algorithm greatly reduces the energy consumption of cluster head node, effective balances the load of network and prolongs the network life time, simultaneously can ensure quality of the reconstructed image. While this paper proves the superiority and the feasibility of the proposed algorithm, there is still a lot of work worthy of in-depth research. Possible direction of future research is to utilize multi-path routing to enhance the performance of distributed image compression and eliminate blocking effect. 6. References
[1] F. Akyildiz, T. Melodia, K. R. Chowdhury. A survey on wireless multimedia sensor networks [J]. Computer Networks, 2007, 51(9): 92-960. [2] Huaming Wu, A. Abouzeid. A Energy efficient distributed image compression in resource-constrained multihop wireless networks [J]. Computer Communications. 2005, 28(14): 1658-1668. [3] Skodras A, Christopoulos C. The JPEG2000 still image compression standard [J]. Signal Processing Magazine, 2001, 18(5): 36-58. [4] R. Dai and I. F. Akyildiz. A spatial correlation model for visual information in wireless multimedia sensor networks. IEEE Transactions on Multimedia [C], 2009, 11(6):1148-1159. [5] Chung-Jr Lian, Kuan-Fu Chen. Analysis and architecture design of block-coding engine for EBCOT in JPEG2000 [J]. IEEE Transactions on Circuits and Systems for Video Technology, 2003, 3(13): 219-230. [6] Yang Dongyong, Gu Dongyuan, An Indoor Location Algorithm Basedon RSSI-Similarity Degree. Chinese Journal of Sensors and Actuators. 2009, 22(2):264-268. [7] S. Misra, M. Reisslein and G. Xue, "A Survey of Multimedia Streaming in Wireless Sensor Networks", IEEE Communications Surveys & Tutorials, Vol. 10, No.4, Fourth Quarter 2008. [8] R. Puri, A. Majumdar and K. Ramchandran, "PRISM: A Video Coding Paradigm with Motion Estimation at the Decoder", IEEE Transactions on Image Processing, Vol. 16, No. 10, October 2007. [9] R. C. Gonzalez, R. E. Woods, and S. L. Eddins. Digital Image Processing Using MATLAB. Prentice Hall, 2004.