Andriod平台下手机防盗软件的设计与实现

Low-complexity and energy efficient image compression scheme for wireless sensor networksQin Lu a,*,Wusheng Luo a ,Jidong Wang a ,Bo Chen baDepartment of Instrument Science and Technology,College of Mechatronics Engineering and Automation,National University of Defense Technology,Changsha 410073,China bDepartment of Mathematic and System Science,College of Science,National University of Defense Technology,Changsha 410073,Chinaa r t i c l e i n f o Available online 29May 2008Keywords:Image compressionComputational complexity Energy consumption WSNsa b s t r a c tCurrently most energy-constrained wireless sensor networks are designed with the object of minimizing the communication power at the cost of more computation.To achieve high com-pression efficiency,the main image compression algorithms used in wireless sensor net-works are the high-complexity,state-of-the-art image compression standards,such as JPEG2000.These algorithms require complex hardware and make the energy consumption for computation comparable to communication energy dissipation.To reduce the hardware cost and the energy consumption of the sensor network,a low-complexity and energy effi-cient image compression scheme is proposed.The compression algorithm in the proposed scheme greatly lowers the computational complexity and reduces the required memory,while it still achieves required PSNR.The proposed implementation scheme of the image compression algorithm overcomes the computation and energy limitation of individual nodes by sharing the processing of tasks.And,it applies transmission range adjustment to save communication energy dissipation.Performance of the proposed scheme is investigated with respect to image quality and energy consumption.Simulation results show that it greatly prolongs the lifetime of the network under a specific image quality requirement.Ó2008Elsevier B.V.All rights reserved.1.IntroductionMany of the potential WSN applications such as target tracking,process control,source localization,discovering and following rare animal species,controlling the vehicle traffic on highways and railways necessitate efficient im-age communication in sensor networks.Since in these im-age-based applications,resources-constrained sensor nodes need to capture,compress and transmit a large amount of data,processing and communication efficiency of the compression algorithms is clearly a design con-straint,which needs to be carefully addressed [1].Recently a number of research efforts are under way to address the issue of image compression and transmissionin sensor networks.Pradhan et al.proposed a distributed coding framework to realize the coding gain of correlated data from the Slepian–Wolf coding theorem in informa-tion theory [2].Wagner et al.proposed another distrib-uted image compression scheme [3]by sending the low-resolution overlapped areas to the receiver and using super-resolution recovery techniques to reconstruct them.Min Wu et al.proposed a novel collaborative image trans-mission scheme [4].This scheme firstly employs a shape matching method along the route to find out the maximal overlap between images.Then the original image and the difference between another image and it are coded to transmit instead of transmitting two individual images independently.These works are designed with the object of minimizing the communication power at the cost of more computation.Except the distributed image compression for multiple correlated images based on1389-1286/$-see front matter Ó2008Elsevier B.V.All rights reserved.doi:10.1016/net.2008.05.006*Corresponding author.E-mail address:freda0126@ (Q.Lu).Computer Networks 52(2008)2594–2603Contents lists available at ScienceDirectComputer Networksj o u r n a l h o m e p a g e :w w w.e l s e v i e r.c o m /l o c a t e /c om n e tthe Slepian–Wolf coding theorem,few schemes consider the computational complexity of the image compression algorithms.The high-complexity state-of-the-art image compression standards,such as JPEG2000,are still the main methods they used.However,these image compres-sion algorithms are not suitable for resource-constrained wireless sensor networks because they require complex hardware and make the energy consumption for compu-tation comparable to communication energy dissipation. Therefore,a low-complexity and high efficient image compression algorithm should be designed in order to re-duce the hardware cost and the energy consumption of the sensor network.To prolong the lifetime of the network,the implementa-tion scheme of image compression algorithm should also be considered in the design.Traditionally,source coding is implemented at the source in order to reduce the num-ber of bits transmitted thus reducing the communication energy dissipation.However,this approach is not always energy efficient for image-based applications.Centralized image compression which is used at the source nodes may sometimes limit the lifetime of the network.In this paper,we present a low-complexity and energy efficient image compression scheme that is suitable for re-source-constrained wireless sensor networks.The key fea-tures of this proposed scheme are:Lapped biorthogonal transform(LBT)is used in image compression instead of discrete cosine transform(DCT) or discrete wavelet transform(DWT).Compared with the DCT-based methods,the proposed algorithm improves coding efficiency by taking into account inter-block spatial correlation and solves the problem of block-ing pared with the DWT-based ones,the proposed algorithm greatly lowers the complexity of computation and reduces the required memory,while it still achieves required peak signal-to-noise ratio (PSNR).What is more,it is very suitable for distributed implementation in the sensor network.Golomb+Multiple Quantization(MQ)coders are used in image compression instead of Huffman coding or arithmetic coding,so the computational complexity is significantly reduced.Further,the memory requirement is minimized because the proposed algorithm doses not require any statistical table or list when coding.A distributed implementation scheme of the LBT-basedimage compression algorithm is proposed based on a clustering architecture.It overcomes the computation and energy limitation of individual nodes by sharing the processing of tasks.Both computational and com-munication energy consumption are considered.This greatly prolongs the lifetime of the wireless sensor net-work under a specific image quality requirement.The rest of the paper is organized as follows.In Section 2,we briefly review related work.The algorithm of LBT and its corresponding encoding method is introduced in Sec-tion3.Section4proposes the distributed implementation scheme of the LBT-based image compression algorithm. Simulations of proposed compression algorithm and implementation scheme are presented in Section5.We conclude the paper in Section6.2.Related workTo our knowledge,low-complexity and energy efficient compression schemes for single image in wireless sensor networks have not been studied in the literature.However, our work has been inspired by a variety of related research efforts.We describe some of the ideas and basic concepts below.Nowadays transform-based methods are still the most popular lossy image compression methods which mainly involve DCT-based methods and DWT-based ones.DCT-based methods have many fast algorithms with low-com-plexity and low-memory but often cause boring blocking artifacts in low bit rate.DWT overcomes this innately for its ability to keep track of time and frequency simulta-neously and represent local features better.JPEG2000is a representative DWT-based compression method.It is not appropriate for WSNs because its transform(CDF9/7)and coding process(EBCOT)are both of high computational complexity and need much memory.To overcome the problems of DWT,Chrysafis and Zhang presented line-based wavelet transform[5]and strip-based wavelet transform[6]separately.The memory needed in these two algorithms is only related with the level of transform and the width of the image but not the height.However, the wavelet coefficients used in[5,6]are stillfloating point numbers and the computation is not reduced.If we substi-tute CDF9/7with binary wavelet transform[7]which needs less computation,the PSNR of compressed image will decrease about1dB.In resourced-constrained WSNs,Huaming Wu et al.no-ticed the high energy consumption of JPEG2000first[7]. They distributed the workload of wavelet transform to sev-eral groups of nodes along the path from the source to the destination by using the concept of parallel distributed computing theory.Our work is inspired by this idea.How-ever,it is worth noting that JPEG2000is not suitable for distributed implementation in the sensor network because the processing of tiles independently leads to a rate-distor-tion loss and blocking artifacts as the number of tiles in-creases or bit rate lowers.Wefind that the lapped biorthogonal transform(LBT) [8]can solve the problems above.LBT is improved DCT, and costs about half of computation of CDF9/7wavelet and needs only1/15of the memory.Moreover,LBT has the following advantages.(1)In analysis process,the corre-lation between adjacent blocks is substantially reduced,so the energy of the image is concentrated better.(2)In the synthesis process,thefilter functions of LBT decay smoothly to zero at their boundaries,which can reduce the block effect efficiently.(3)It is suitable for the distrib-uted implementation in the WSN.In the quest for higher compression efficiency,arithme-tic coding has been applied to DCT and DWT.There are sev-eral representatives of such state-of-the-art coders,such as EZDCT[9],EZW[10],SPIHT[11]and EBCOT[12].These coders provide very high compression efficiency.TheQ.Lu et al./Computer Networks52(2008)2594–26032595disadvantage of these coders is higher computational com-plexity and additional memory requirement.The work in [13]proposed a low-complexity and low-memory entropy coder which combined zerotree coding with quantization. All quantization steps are completed before the zerotree coding and no list is needed.A Golomb codec is used when coding the quantized coefficients.As we know,the distri-bution of quantized coefficients of LBT is similar to DWTs. Therefore,the coder used in the proposed image compres-sion algorithm is similar to the coder in[13].Here,it should be pointed out that categorizing is performed after the zerotree coding,and the ability of coding the coeffi-cients of zerotree codec is neglected in[13].In practice, the signal put into the Golomb codec does not follow a geo-metric distribution well.Therefore,we apply an MQ codec after the Golomb codec for further coding.In our previous work,we have used this LBT-based algo-rithm in the image compression device for satellite which is also strictly restricted in space,CPU memory resources and energy consumption[14].The design of the algorithm and its implementation scheme in WSNs has not been studied before.3.Image compression algorithm based on LBTIn WSNs,we need an image compression method with low complexity as well as good image pped bior-thogonal transform(LBT)is a good candidate for this requirement.In this section,we present a lifting-based fast binary algorithm of LBT and a low-complexity zerotree coding method.3.1.Lifting-based fast binary algorithm of LBTThrough a series of elementary matrix manipulations,a large class of lapped transform(LT)can be constructed as the pre-processing of DCT inputs in the time domain or the post-processing of DCT coefficients in the frequency domain[15].Both classes of algorithms share one common goal:to eliminate or reduce the severity of coding artifacts in the reconstructed signal.By analyzing the structures of these two classes of algorithms,we use time-domain pre-processing of DCT in this paper which is attractive for the distributed implementation in WSNs.The structure is de-picted in Fig.1a.Fast LT can be constructed in polyphase form from com-ponents with a fast algorithm for DCT.The analysis poly-phase matrix of type-II fast lapped orthogonal transform (LOT)[16]can be written as:E IIðzÞ¼D M C IIMI00zI0II0P;ð1ÞwhereP¼1I JJÀII00VI JJÀIV¼JðC IIM=2ÞT C IVM=2JD M¼diag1;À1;1;À1;...f g:ð2ÞThe matrix V controls pre-filtering and contains all of the degrees of freedom in the structure.By inserting aninvertible diagonal matrix S betweenðC IIM=2ÞT and C IVM=2,V can be represented in the single-value decomposition form:V¼JC IIM=2T SC IVM=2J;ð3ÞS is limited to diag f s;1;...;1g,where s is a scaling factor.A good value of s for a smooth image model is8=5which yields good biothogonal solutions.It is well known that a plane rotation can be decom-posed into three lifting steps.This can be written in matrix formas:Fig.1.Implementationflow of one-dimension binary. 2596Q.Lu et al./Computer Networks52(2008)2594–2603cos h Àsin h sin hcos h¼10tan h 21"#1Àsin h 0110tan h21"#:ð4ÞSo,we can replace each butterfly in the pre-filter P by theunnormalized Haar matrix asP ¼1I J 1J ÀI"#I 00VIJ 12J À12I "#:ð5ÞBecause the current consumption of the multiplyinstruction in DSPs is far greater than the current con-sumption of other instructions,a lifting scheme is used in the implementation of transform process.All coefficients are approximated by the fraction of integral numerator and dyadic denominator so only integer additions and shifts but no float multiplications are needed;this is called a binary transform and is quite suitable for hardware implementation.Fig.1b shows the structure of the binary LBT pre-processing.In our image compression algorithm,an eight-point type-II DCT [17]is used.The structure of the binary DCT-II is illustrated in Fig.1c.The computations needed by LBT,DCT and DWT (CDF9/7)are listed in Table 1.It is seen that the computation of binary LBT is 62%less than that of binary DWT,while its performance is better than that of binary DWT in the experiments,which will be shown in Section 5.3.2.Low-complexity zerotree codingSensor nodes in WSNs are limited in energy,memory and computational power.Therefore,besides high com-pression efficiency,the coders for image compression inwireless sensor networks should be of low complexity and low memory.The distribution of quantized coefficients of LBT is sim-ilar to DWT which has a zerotree structure.The zeortree structure of 8Â8LBT is illustrated in Fig.2.The element in every pane is a quantized coefficient of LBT.The energy of the coefficients,i.e.,the absolute value,decreases from top left corner to bottom right.Therefore,separating the entire block into 10sub bands by the solid lines,the LBT coefficients can be classified by the sub band that they be-long to and then coded separately.Denoting X as the quantized 8Â8blocks of coefficients,X ði ;j Þ;i ;j ¼0;...;7are the quantized coefficients.C ðX ði ;j ÞÞare the children of X ði ;j Þ,and D ðX ði ;j ÞÞare the descendants of X ði ;j Þexcept C ðX ði ;j ÞÞ.The detail steps of coding are given below:1.Code X ð0;0Þby Golomb +MQ codec;2.Code x ¼X ð0;0Þwith zerotree codec;2.1.If x has no children,end;2.2.If all the children of x is zero,output 0;Otherwise go to (2.3);2.3.Output 1,and code every children of x byGolomb +MQ codec;2.3.1.If all the coefficients in D ðx Þis 0,output 0;Otherwise go to (2.3.2);2.3.2.Output 1,and code the coefficients in C ðx Þwith zerotree codec separately.where the Golomb +MQ codec is Denote input as X ði ;j Þpute C ðX ði ;j ÞÞ¼N ;2N À1X ði ;j Þj j 2N ;2.Get the sub band B that X ði ;j Þbelongs to.Send ð0;B Þinto MQ codec N times and send ð1;B Þinto MQ codec one time;3.If X ði ;j Þis positive,output 1;Otherwise output 0;4.Output the lower C ðX ði ;j ÞÞÀ1bits of X ði ;j Þ.To reduce computation,the further compression of the position and sign information [18]is not considered here.In this improved zerotree coding algorithm,all quantizationisFig.2.Zerotree structure of 8Â8LBT quantized coefficients.Table 1Computation cost per bitFloat DCTFloat CDF9/7Binary DCT Binary CDF9/7Binary LBT Integer+007.7534.113Integer shift 00 3.523.6 6.7Float+7.2510.5000Float Â3.257.9Q.Lu et al./Computer Networks 52(2008)2594–26032597completed before zerotree coding and no list is needed.Thus it needs little memory and is straightforward for both soft-ware and hardware implementation.Here,we call it LZC. 4.Image compression schemeThe goal of presenting a low-complexity and low-mem-ory image compression algorithm based on LBT is to realize the image compression scheme in resource-constrained WSNs.We now propose a distributed implementation scheme of this image compression algorithm.4.1.System modelIn our study,we use the wireless communication en-ergy model proposed in[19].The energy consumed in transmission per bit isE TX¼E elecþe fs d2;d<d0E elecþe mp d4;d P d0(ð6Þand the energy consumed in reception per bit isE RX¼E elec;ð7Þwhere E elec is the energy consumed by the circuit per bit,d is the distance between the wireless transmitter and the receiver,and e fs d2or e mp d4is the amplifier energy that de-pends on the transmitter amplifier model.The energy consumed in image compression per bit is E cp¼2ðE preþE DCTÞþE code;ð8Þwhere E pre is the energy dissipated for one-dimensional bin-ary LBT pre-processing.E DCT is the energy dissipated for one-dimensional binary DCT.E code is the energy spent in coding.Noting r as the compression ratio,then if E cpþE TX= r>E TX,it’s better for the camera sensor to send the raw data directly than to compress it before sending.work structureAs each sensor node is power-constrained,the lifetime of WSNs is limited.Efficiently organizing sensor nodes into clusters is useful in reducing energy consumption.We con-sider the image compression scheme in the clustered wire-less sensor network because clustering allows for scalability of MAC and routing.What is more,cluster heads can serve as fusion points for the aggregation of data,so that the amount of data that is actually transmitted to the base station is reduced.Although data aggregation is not discussed in this paper,it should be considered when we design the image compression scheme.A typical cluster-based method is LEACH[20].In LEACH, the entire network is divided into non-overlapping clus-ters.There is a cluster head among each cluster.Instead of transmitting the data to the base station directly,the sensors send their data to the cluster head which relays the data to the global base station.Many LEACH-like algo-rithms,such as DEEC[21]and LEACH-B[22],have recently been proposed to improve the performance of LEACH re-cently.In DEEC,the cluster heads are elected by a probabil-ity based on the radio between the residual energy of each node and the average of the network.Because we are not aiming at the performance of the clustering schemes,the proposed scheme is designed based on LEACH in this paper.Two types of nodes with different functionality are used.One is the camera node;the other is the scalar sen-sor.The nodes are assumed to have power control features so as to adjust their transmit power.By using LEACH,all the nodes commonly run in a single hop homogeneous network(Fig.3left).The cluster-head role is rotated ran-domly and periodically to distribute the energy consump-tion evenly among all sensors in the network.The camera nodes should adjust their transmit power to communicate with the cluster head.If the cluster head is far from the camera node,the communication energy consumption will be high.Hence,this architecture is not energy efficient for the camera nodes.To prolong the lifetime of the network,both computa-tional and communication energy consumption should be considered.To reduce communication energy of the cam-era nodes,we propose an architecture in which the camera node does not take part in the clustering process.It adjusts its transmission range only to keep its neighborsnearlyFig.3.(Left)all the nodes commonly run in a single hop homogeneous network;(right)the camera node does not take part in the cluster and adjusts transmission range only to keep its neighbors nearly constant.2598Q.Lu et al./Computer Networks52(2008)2594–2603constant during the cluster head selection phase and trans-mits data to the neighbors during the steady-state phase. Thus,the communication energy consumption of the cam-era node is minimized.4.3.Partition schemeThe purpose of distributed implementation of the image compression algorithm is to reduce the burden ofenergy Fig.4.(a)Normal row-to-column computing scheme;(b)proposed partition scheme for distributed implementation inWSNs.Fig.5.Distributed implementation of LBT-based image compression algorithm in a cluster.Q.Lu et al./Computer Networks52(2008)2594–26032599consumption in the camera sensor and maximize the net-work lifetime.Therefore,we propose a partition scheme of the image compression algorithm which is different from the row-to-column way in normal parallel distributed computing(Fig.4a).The scheme is illustrated in Fig.4b.First,12rows data are read into the buffer and the back8rows do the LBT pre-processing on all columns.Second,the former8rows do DCT on all columns.Third,the former8rows do the LBT pre-processing and DCT on all rows.Once the former 8rows complete their2D LBT,they can be coded in8Â8 blocks and sent out directly.Then the buffer reads the next 8rows in and repeats the same process with the4rows left.In this partition scheme,only12W(image width)pixels need to be buffered at one time.It can be seen that the processes4,2,and5all deal with the former8rows until the compression of the8rows data complete.So these tasks can be distributed to other nodes in the cluster.4.4.Distributed image compressionDistributed implementation of the LBT-based image compression algorithm in a cluster is shown in Fig.5.When the camera node takes a picture,it should send it to the cluster head.During the transmission,the image is com-pressed.The compression scheme follows the steps below: Step1:The camera node announces this message to one of its neighbors S.Step2:S sends a request to the head in its cluster.Step3:The cluster head chooses some idle nodes ðP1ÀP4Þin the cluster which have the energyabove a threshold and send an OK message backto S.Step4:S sends an OK message back to the camera node.Table2The PSNR of different algorithms(dB)Rate (bpp)BinaryLBT+LZCBinaryDCT+LZCBinaryDWT+LZCFloat CDF9/7+SPIHT138.7838.1538.6339.380.536.1335.4136.0536.200.2533.0832.2932.9633.12pression of Lena image.2600Q.Lu et al./Computer Networks52(2008)2594–2603Step5:The camera node sends8rows data(12rows data in thefirst time)to S.S does the process3andsends the former8rows to a chosen node P whichis now idle.Then S goes to Step4.Step6:Each chosennode P does the processes4,2,and5, and sends the compressed data to the Cluster head.The time of doing step6is much more than the time of doing step5,hence the compression scheme needs many Ps to compute at the same time.In this scheme,12W pixels need to be buffered in the processor of S and8W pixels need to be buffered at each processor P.While in5-level decomposition with CDF9/7, even if the line-based transform method is used,about ð25þ1þ2À5À3ÞÂ3W%183W pixels must be buffered. The buffer that LBT needs is only1/15of that of the DWT.5.Performance evaluationThis section compares the performance of the proposed image compression algorithm with other image compres-sion algorithms with respect to two metrics:image quality and the network lifetime.5.1.Image qualityIn this analysis we focus on the performance of not only the algorithm itself but also its distributed implementation in WSNs.The image Lena of size512Â512pixels with8 bits per pixel is used in this evaluation.To evaluate the performance of the proposed image compression algorithm,we compare four algorithms:the proposed one,binary DCT+LZC,5-level binary DWT+ LZC and5-level Float CDF9/7+SPIHT.Table2gives the PSNR of these algorithms.It can be observed that binary LBT outperforms binary DCT.And the PSNR of recon-structed image compressed by the binary LBT-based algo-rithm is higher than the binary DWT-based one. Compared with Float CDF9/7+SPIHT,the binary LBT-based algorithm is also competitive.Because of using a binary transform and low-complexity coding algorithm,the PSNR decreased0.24dB on the average.In the case of distributed implementation in WSNs, nodes are extremely limited in memory.Therefore we compare the LBT-based distributed method with the dis-tributed JPEG2000method in[7]under the same memory requirement condition.In the proposed LBT-based distrib-uted method,8Â512pixels of the image is processed by the nodes at a time.In the distributed JPEG2000method, the image is partitioned into tiles,and then each tile is sent to a node to do JPEG2000algorithm independently.To let the image size processed by one node be equal for these two methods,here we consider the distributed JPEG2000 method with64Â64pixels tile.Fig.6shows the image quality with0.1bits per pixel(bpp)and0.25bpp bit rates. We can see that the image quality of the LBT-based method is much better than the distributed JPEG2000one when the bit rate is low.work lifetimeIn this section,we evaluate the performance of the pro-posed distributed image compression scheme using MAT-LAB.We consider seven sensor networks with N¼25;50; 100;250;400;625or900nodes randomly distributed over a500mÂ500mfield.We choose the node which is clos-est to the center of thefield as the camera node.The coor-dinate of the base station is(0,0).Each node is initially given10J of energy.LEACH is used in the simulation and here P opt¼0:1.Fig.7.System lifetime of the three schemes.Q.Lu et al./Computer Networks52(2008)2594–26032601The values of the energy model parameters are chosen as follows:the values of the parameters of the wireless com-munication energy model in (6)and (7)are the typical val-ues E elec ¼50nJ =bit,e fs ¼10pJ =bit =m 2,e mp ¼0:0013pJ =bit =m 4and d 0¼87m.The energy consumption of the pro-posed image compression algorithm is evaluated in a StrongARM SA1100processor.The experiment is done on the test image Lena of size 512Â512pixels with 8bits per pixel.The bit rate is 0.25bpp.From the experiment,the val-ues of the parameters of the computation energy model are E pre ¼15nJ =bit,E DCT ¼20nJ =bit and E code ¼90nJ =bit.To evaluate the performance of our proposed scheme,we compare the following three schemes:Scheme (A):The camera node acts as a member of the nearest cluster and compresses the images before send-ing them to the cluster head.Scheme (B):The camera node acts as a member of the nearest cluster and sends the raw images to the cluster head.Then the cluster head compresses the images and sends the compressed data to the base station.Scheme (C):The proposed distributed image compres-sion scheme in the proposed network architecture.We first let the camera node adjust its transmission range to keep one neighbor alive in its communication ra-dius at the beginning of each round.Five images are in-tended to be sent out during every steady-state phase.The simulation is stopped if the camera node depletes its energy.The system lifetime is measured by the number of images which are successfully received by the base sta-tion.Fig.7shows the system lifetime of the three schemes.It is observed that the proposed scheme results in a much longer lifetime since the camera node shares the computa-tional workload with other nodes and adjusts its transmis-sion range to save communication energy consumption.In the design of WSNs both energy efficiency and QoS support should be considered.In the proposed scheme,we can let the camera node adjust its transmission range to keep more than one neighbor alive in its communication radius.The cluster head is in charge of measuring the en-ergy and data loss probability of its cluster.And the camera node chooses the best cluster to process the distributed image compression scheme according to the message from the cluster head.Here,we let the camera node choose the cluster according to the residual energy of the cluster head.Fig.8shows the system lifetime of the proposed scheme when the camera node keeps M =1,2or 3neighbors with-in its communication radius.It is observed that in the net-work with a high node density the proposed scheme performs well when the camera node increases its trans-mission range to keep more neighbors alive in its commu-nication radius.6.ConclusionsWe have studied the problem of image compression algorithm and its implementation scheme in WSNs.The de-sign and evaluation of a low complexity and energy efficient image compression scheme is presented.A LBT-based low complexity and low memory image compression algorithm is used.It greatly reduces the hardware cost and the energy consumption of the sensor network.To prolong the lifetime of the network,both computational and communication energy consumption are considered in the implementation scheme.Our technique shares the task of image compres-sion algorithm with several nodes and applies transmis-sion range adjustment to save communication energy dissipation.The proposed scheme is simple to implement.Performance evaluation shows that our technique greatly prolongs the lifetime of the wireless sensor network under a specific image qualityrequirement.Fig.8.System lifetime of the proposed scheme.2602Q.Lu et al./Computer Networks 52(2008)2594–2603。