PERFORMANCE ANALYSIS FOR DCT-BASED CODED IMAGE COMMUNICATION IN WIRELESS MULTIMEDIA SENSOR NETWORKS

Transcription

1 INTERNATIONAL JOURNAL ON SMART SENSING AN INTELLIGENT SYSTEMS VOL. 6, NO. 1, FEBRUARY 013 ERFORMANCE ANALYSIS FOR CT-BASE COE IMAGE COMMUNICATION IN WIRELESS MULTIMEIA SENSOR NETWORKS Xiong Zheyuan 1, Fan Xiaping 1,, Liu Shaoqiang 1, Li Yongzhou 1 and Zhang Huan 1 1 Shool of Informaion Siene and Engineering Cenral Souh Universiy Changsha, China Laboraory of Neworked Sysems Hunan Universiy of Finane and Eonomis Changsha, China s: xiongzhey@16.om; xpfan@mail.su.edu.n; liussqq@16.om; Submied: O. 8, 01 Aeped: Jan. 6, 013 ublished: Feb. 0, 013 Absra- The objeive of his paper is o sudy he image ommuniaion behavior of a wireless video sensor, and analyze is performane under resoure onsrained wireless mulimedia sensor neworks. Energy onsumpion and rae disorion model of CT-based oded image ompression and ransmission is developed. Based on energy onsumpion and rae disorion model proposed, resoure alloaion is opimized wih limied energy and bandwidh. Number of CT oeffiien and sep size of quanizaion able is adaped aording o he aiviy of monioring sene, and hen he energy onsumpion will be minimized. Simulaions resuls are ondued o show he performane of our 10

2 Xiong Zheyuan, Fan Xiaping, Liu Shaoqiang, Li Yongzhou and Zhang Huan, erformane Analysis For CT-Based Coded Image Communiaion in Wireless Mulimedia Sensor Neworks work. The proposed sheme dramaially redues image ompression and ransmission energy onsumpion under expeed image disorion and ransmission rae. Index erms: Wireless mulimedia sensor neworks, image ompression, image ransmission, CT. I. INTROUCTION Wireless mulimedia sensor neworks (WMSN) onsis of iny amera nodes, whih inegrae he image sensor, embedded proessor, and wireless ranseiver. Beause of is unique feaures of rapid deploymen, flexibiliy, low mainenane os, and robusness, WMSN have developed as a new ehnology wih a number of poenial appliaions, ranging from seuriy o monioring o elepresene [1]. The major ask of WMSN is image ommuniaion, whih is really a hallenge for resoure onsrain wireless sensors. The ulimae goal of image ommuniaion design for WMSN is o opimize he performane under limied resoures. In radiional video ommuniaion appliaions, he major resoure onsrain is in he form of ransmission bandwidh or sorage spae, whih deermines he oupu bi rae of a video enoder. Compared o radiional ommuniaion sysems, WMSN operaes under a se of unique resoure onsrains, inluding limied energy supply, on-board ompuaional apabiliy, and ransmission bandwidh. Therefore, here is a need o exend he radiional R- analysis by onsidering hese new resoure onsrains in he WMSN. The objeive of his sudy is o analyze he impa of resoure onsrains on he performane limi of a wireless video sensor. Image ommuniaion is he dominan power-onsuming operaion in wireless video sensors over WMSN, whih inludes image ompression and ransmission. For he resoure onsrains in WMSN, he effe of image enoding is wo-fold. Firs, effiien image ompression signifianly redues he amoun of he video daa o be ransmied, whih in urn saves a signifian amoun of energy in image ransmission. Seond, more effiien image ompression ofen requires higher ompuaional omplexiy and larger power onsumpion in ompuing. These wo onfliing effes imply ha in praial sysem design here is always a radeoff among he bandwidh R, power onsumpion, and video qualiy. Therefore, o find he bes rade-off soluion, we need 11

3 INTERNATIONAL JOURNAL ON SMART SENSING AN INTELLIGENT SYSTEMS VOL. 6, NO. 1, FEBRUARY 013 o develop an analyi framework o model he performane of he image ommuniaion in WMSN. Image ompression is lossy, whih no only inrodues disorion in he oded soure in he form of quanizaion errors, bu also makes he daa sream highly sensiive o bi errors when ransmied over wireless hannels []. Video disorion in wireless neworks is mosly aused by lossy soure oding, ransmission errors originaed by hannel fading, buffer overflows and playou deadline misses. Sine reharging for wireless video sensors is no praial, minimizing he energy onsumpion of wireless video sensors under resoure onsrains, whih inludes expeed disorion of images and ransmission rae, will be ruial. For image enoding on video sensor and sreaming over WMSN, i is needed o analysis Rae-disorion (R-) under energy onsrains. In energy-aware image enoding, he oding disorion is no only a funion of he enoding bi rae as in he radiional R- analysis, bu also a funion of he power onsumpion. udlewski e al. propose a rae onrol sheme based on boh analyial and empirial models, along wih a join video and hannel enoder rae alloaion sheme [3]. The sheme joinly regulaes he end-o-end daa rae, he video qualiy, and he srengh of he hannel oding a he physial layer o minimize he disorion of he reeived video, and finds he ombinaion of video and hannel enoding ha resuls in he bes qualiy video a he reeiver. Zhihai He e al. analyze he -R- performane and develop an adapive sheme o esimae he -R- model parameers and perform online energy opimizaion and onrol for real-ime video ompression [4]. Zhao Sun e al. develop an online omplexiy onrol and energy minimizaion sheme for real-ime video enoding over porable devies [5]. They inrodue a se of parameers o onrol he ompuaional omplexiy and energy onsumpion of he video enoder. Based on neural nework ompuaions, he sheme deermines he omplexiy onrol parameers o maximize he overall video enoding performane under energy onsrains. Zhenhua Tang e al. presen an energy-minimized resoure alloaion sheme, whih applies UE of hannel oding and UT joinly [6]. In his sheme, unequal ransmission proeions are adapively assigned o various pars of ompressed images aording o heir differen sensiiviies o ransmission errors. So far, exising mehods and algorihms have been foused on enoder omplexiy and energy onsumpion reduion hrough heurisi adapaion or onrol insead of sysemai energy 1

4 Xiong Zheyuan, Fan Xiaping, Liu Shaoqiang, Li Yongzhou and Zhang Huan, erformane Analysis For CT-Based Coded Image Communiaion in Wireless Mulimedia Sensor Neworks opimizaion. This is beause hey lak an analyi model o haraerize he opimum rade-off beween energy onsumpion and enoding performane. In addiion, even wih exising energy saving ehnologies, he operaional lifeime of porable video eleronis sill remains very shor whih has beome one of he bigges impedimens o our ehnology fuure. Therefore, i is imporan o develop new energy minimizaion approahes for WMSN. In his work, we fous on energy onsumpion analysis and rae disorion analysis of image ompression and image ransmission in WMSN sine hese wo onsume a major porion of he oal energy while he energy onsumpion of he remaining omponens depends on speifi sysem design and is no easily onrollable from an image ompression perspeive. Then a resoure alloaion opimizaion sheme is proposed aording o he analysis. II. ENERGY CONSUMTION ANALYSIS 1. Energy Consumpion of Image Compression Typially, visual informaion needs o be ompressed using one of he well known ompression sandards suh as JEG or JEG000 o save ransmission bandwidh or sorage spae. Unforunaely, hese sandards are ompuaional inensive and energy onsuming. Sine he limiaions wih respe o energy resoures and on-board proessing apabiliy, we have o adap a lighweigh image ompression algorihm o he requiremen of WMSN [7]. The JEG algorihm is uilized for he work presened here. While he JEG000 sandard offers some signifian advanages over he JEG, i is muh less widely deployed and also requires full-frame proessing whih an be hallenging in ligh of memory limiaions [8]. Typial video enoders, inluding video and image oding sandards, suh as MEG-, H.63, MEG-4, and JEG employ CT enoding sheme. They have he following hree major enoding modules: CT, quanizaion, enropy enoding of he quanized CT oeffiiens [9]. 13

5 INTERNATIONAL JOURNAL ON SMART SENSING AN INTELLIGENT SYSTEMS VOL. 6, NO. 1, FEBRUARY 013 Le Figure 1. JEG enoding be he required energy o ompress he image daa for JEG sages. formulaed as follows: Where d, q and enropy oding sages, respeively. an be d q enropy (1) enropy represen he energies onsumed a CT, quanizaion, and A he enoder, soure image samples are grouped ino 8 8 bloks, shifed from unsigned inegers wih range [0, 1] 1 1 o signed inegers wih range[, 1], and inpu o he Forward CT (FCT). A he deoder, he Inverse CT (ICT) oupus 8 8 sample bloks o form he reonsrued image. The following equaions are he idealized mahemaial definiions of he FCT: k 1 k 1 1 (x 1) u (x 1) v F( u, v) C( u) C( v)[ f ( x, y)*os os ] k k k x 0 y 0 () Where: C( u), C( v) 1/ for u, v 0; C( u), C( v) 1 oherwise. Where uvare, he disree frequeny variables suh ha 0 u, v k 1, k is he blok size. f ( x, y ) is he grey level of one pixel a (x, y) in he N N image, where 0 x, y k 1; F( u, v) is he oeffiien of he poin ( uv, ). Eah k k blok neessiaes k log k mulipliaions and 3k log k k k addiions for row-olumn mehod. Therefore he energy dissipaed for one blok is ( log ) (3 log ) k k pmul k k k k p add, where mul p and p add represen he energy onsumpion for mulipliaion and addiion insruions. For an image of size N N having N ( ) k bloks, he oal energy dissipaed is: 14

6 Xiong Zheyuan, Fan Xiaping, Liu Shaoqiang, Li Yongzhou and Zhang Huan, erformane Analysis For CT-Based Coded Image Communiaion in Wireless Mulimedia Sensor Neworks N N p N k p N k N p k k d ( ) d ( log ) mul (3 log ) add The proessing energy by he quanizaion sage an be expressed as a funion of F( u, v ) and Q( u, v ), where Q( u, v ) is a oeffiien of he luminane quanizaion marix Q defined for k 8, 0 u, v k 1. Quanizaion is defined as division of eah CT oeffiien by is orresponding quanizer sep size, followed by rounding o he neares ineger: Therefore, here are is k divisions and Q F( u, v) F ( u, v) Ineger Round ( ) (4) Q ( u, v ) (3) k round operaions per blok and he energy dissipaed p q k ( pdiv p r ), where p div and p r are he energy onsumed for division and round insruions, respeively. For one image N N, he oal energy dissipaed in his sage is: N N p k p p N p p k k q ( ) q ( ) ( div r) ( div r) Afer quanizaion, all of he quanized oeffiiens are ordered ino he zigzag sequene. Zigzagging is a simple rearrangemen of he (5) ( k 1) AC oeffiiens ordered in a zigzagging way, from lower o higher frequenies. The zigzagging energy onsumpion on eah blok an be alulaed as: pz ( k 1) p sh, where sh p represens he energy onsumpion of he shif proess. For one inpu image of size N N, where z represens he energy onsumed a zigzagging sage, he oal proessing energy relaed o he zigzagging sage is: N N p k p k k z ( ) z ( ) ( 1) sh Enropy oding an be onsidered as a sep proess. The Baseline sequenial ode uses run lengh enoding and Huffman oding for hese seps. Run lengh enoding of he (6) ( k 1) AC oeffiiens is a rearrangemen of he zigzagged sequene. I onvers he zigzag sequene of quanized oeffiiens ino an inermediae sequene of symbols. Huffman enoding onvers he symbols o a daa sream in whih he symbols no longer have exernally idenifiable boundaries. Enropy oding an be formulaed as follows: enropy rle huf (7) Where rle and huf represen he energy onsumed a RLE and Huffman enoding sages, respeively. Run lengh enoding of he AC oeffiiens an be inerpreed mainly as a 15

7 INTERNATIONAL JOURNAL ON SMART SENSING AN INTELLIGENT SYSTEMS VOL. 6, NO. 1, FEBRUARY 013 rearrangemen of he zigzagged sequene. For Huffman enoding, he oal energy dissipaed per image for boh C and AC oeffiiens.. Energy Consumpion of Image Transmission The image ransmission power is given by Where R ( d n ) R (8) is he energy os ha is needed for suessful ransmission of one daa bi. I depends on ransmission disane and pah loss index. ransmission disane, n is he pah-loss exponen, and R is he ransmission bi rae, d is he is a model parameer relaed o enoding effiieny. Sine he oal power onsumpion is fixed, and, if we inrease, hen dereases. This implies ha less bis an be ransmied beause he ransmission energy is proporional he number of bis o ransmi. Therefore, image disorion inreases. III. RATE ISTORTION ANALYSIS The major ask of wireless video sensor is o ompress image daa and send he ompressed bi sream o he aggregaion and forwarding nodes (AFN). Is performane is measured by he qualiy of videos delivered o he AFN. erformane of image ommuniaion is measured by he qualiy of images delivered o he aggregaion and forwarding nodes, whih is ommonly measured by he end-o-end disorion, denoed by. I onsiss of wo pars: image ompression disorion and ransmission disorion. The image ompression disorion is aused by quanizaion errors during lossy video ompression and is ofen measured by he mean squared error (MSE) beween he original image and he enoder reonsruion, denoed by. The ransmission disorion is aused by ransmission errors, suh as pake loss, and is measured by he MSE beween enoder reonsruion and deoded image a he reeiver, denoed by 1. Image Compression isorion The image ompression disorion is aused by lossy image ompression and is ofen measured by he mean squared error beween he original image and he enoder reonsruion, denoed by [10]. The image ompression bi rae R and disorion. an be onsidered as funions 16

8 Xiong Zheyuan, Fan Xiaping, Liu Shaoqiang, Li Yongzhou and Zhang Huan, erformane Analysis For CT-Based Coded Image Communiaion in Wireless Mulimedia Sensor Neworks of, whih is he perenage of zeros among he quanized CT oeffiiens. R and ommonly reaed as funions of he quanizaion sep size q, and are monoonially inreases wi q. Therefore, mahemaially, R and are also funions of. The disorion model for image ompression is given by: Where (1 ) e (9) ( ) is he variane of image daa and. Image Transmission isorion is a onsan. In wireless video surveillane, wireless video sensor is allowed o reransmi he video pakes if hey are no reeived orrely. However, in WMSN monioring, beause of he sringen delay requiremen, pake reransmissions are infeasible even if he pake is reeived wih errors. These ransmission errors will be passed o he deoder and ause deoding errors and error propagaion o he subsequen frames. In wireless image oding and ransmission, hannel oding is ofen used o orre bi errors in he oded video daa sream. Inreased error resiliene ofen omes a he os of higher bandwidh onsumpion. ue o he limied error orreion apaiy of he hannel deoder, residual bi errors ofen sill exis afer error orreion. When a orruped odeword in he bi sream anno be properly deoded, he video enoder will jump o he nex pake saring wih a resynhronizaion mark and skip all he inermediae bis. This inrodues visible piure disorion a he reeiver. Noe ha a he deoder he urren reonsruion frame orruped by bi errors may be used as he referene for he nex frame. In his way, he hannel disorion ofen severely degrades he video presenaion qualiy. The ransmission disorion is aused by ransmission errors, and is measured by he MSE beween enoder reonsruion and deoded image a he reeiver, denoed by. In WMSN monioring appliaions, beause of he sringen delay requiremen, pake reransmissions are infeasible even if he pake is reeived wih errors. The video enoder will jump o he nex pake and skip all he orruped bis. The ransmission errors will ause deoding errors and error propagaion o he subsequen frames [11]. The average hannel disorion aused by pake loss is proporional o he mean frame differene. Image ransmission disorion aused by pake loss as pake reransmissions are infeasible, is given by 17

9 INTERNATIONAL JOURNAL ON SMART SENSING AN INTELLIGENT SYSTEMS VOL. 6, NO. 1, FEBRUARY 013 a p F d () i 1 b b 1 p (10) Le p be he pake loss raio and be he average fraion of bloks in he image. a and b are model parameers, a is a onsan ha an be regarded as he energy loss raio of he enoder filer, b is a onsan ha desribing he obje moion randomness of he monioring sene. F () i represens he MSE beween he urren ompressed frame and he previous frame, d F ( i) af ( i ). F () i is he MSE beween he urren frame and he previous frame. Fd () i is d d d he average frame differene Fd () i of he oal image sequene. x M, y N 1 F ( i) f ( x, y) f ( x, y) d i i M N x 1, y 1 1 (11) IV. RESOURCE ALLOCATION OTIMIZATION Based on he analysis previously, energy onsumpion and rae-disorion of image ompression and ransmission an be opimized. In WMSN monioring appliaion, a video sene ofen experienes a series of evens wih ime varying. Therefore, he daa volume of eah image frame o be enoded by wireless video sensor is ofen nonsaionary. The enoder may rea he image sequenes differenly and performs opimum resoure alloaion among hese images o minimize he overall enoding power onsumpion. To saving energy under resoure onsrains, enoder mus be adaped o saisfy he aual siuaion in he monioring sene. In WMSN monioring appliaion, a video sene ofen experienes a series of evens wih ime varying. Therefore, he daa volume of eah image frame o be enoded by wireless video sensor is ofen nonsaionary. The enoder may rea he image sequenes differenly and be adaped o saisfy he aual siuaion in he monioring sene o minimize he enoding power onsumpion [1]. Sine he oal power onsumpion is fixed, and, where is he power onsumpion of ompression and is he power onsumpion of ransmission. If inreases, hen dereases. This implies ha less bis an be ransmied beause he ransmission energy is proporional he number of bis o ransmi. Therefore, image disorion inreases. If dereases, hen inreases. The enoder will generae more bis o ahieve he arge 18

10 Xiong Zheyuan, Fan Xiaping, Liu Shaoqiang, Li Yongzhou and Zhang Huan, erformane Analysis For CT-Based Coded Image Communiaion in Wireless Mulimedia Sensor Neworks disorion. This leads o an energy radeoff beween image enoding and wireless ransmission. This energy radeoff an onepually desribed by he following minimizaion problem: min ( d q enropy ) (1) In CT based image ompression, loss is inrodued boh in quanizaion proess and CT whih, while in heory lossless, in praie inrodues loss due o finie preision. And he wo seps onsume mos of he ompression power, almos 70%. Therefore, opimizaion of image ompression energy onsumpion fouses on CT and quanizaion [13]. Besides image oding, hannel oding is needed o orre he bi errors ha inrodued o he video daa sream. In his ase, he overall bi rae is divided ino wo pars: image oding bi rae R and hannel oding bi rae R. To minimize he overall image disorion, he parameers of he image enoder and he hannel odes need o be adjused for differen inpu image sequenes and differen hannel ondiions. As he oal bi rae is resried by he limied wireless hannel bandwidh, bi alloaion beween soure and hannel oding mus be opimized. The enoder seings an be adjused o ahieve he alloaed bi budge. Image ompression disorion and image ransmission disorion are unorrelaed wih eah oher, and hey an be esimaed separaely. and soure oding bi rae ransmission energy onsumpion is a funion of image enoding energy onsumpion R, whih is denoed by ( R, ). is a funion of he and ransmission disane d, whih is denoed by (, d ). Therefore, he resoure alloaion and performane opimizaion of wireless video sensor an be formulaed as follows: min ( R, ) (, d ) s.. (13) R {,, } As he oal bi rae is resried by he limied wireless hannel bandwidh, bi alloaion beween soure and hannel oding mus be opimized. The enoder seings an be adjused o ahieve he alloaed bi budge. ue o varying hannel ondiions and inpu images, he sysem parameers are required o be opimized joinly in ime o mainain he oal sysem energy minimum under he onsrains of he expeed image disorion and ransmission rae. Using he energy onsumpion and disorion models of image ompression and image ransmission and he opimal onfiguraion of he onrol parameers, he wireless video sensor is able o ahieve he opimized power onsumpion 19

11 INTERNATIONAL JOURNAL ON SMART SENSING AN INTELLIGENT SYSTEMS VOL. 6, NO. 1, FEBRUARY 013 and image disorion. The following seps should be performed o opimize he resoure alloaion: (1) Colle sene aiviy saisis and lassify he sene. Sene aiviy is oarsely divided ino sai and moion by moion deeion. () Enoder omplexiy onrol. (3) Energy onsumpion and disorion models parameers updae. While here is no obje moion in he monioring sene, he inpu of he enoder is saionary. Only low frequeny oeffiiens of CT bloks will be proessed o minimize he number of operaions a eah sage of he ompression sheme. runing algorihm alulaing k k par of he CT oeffiiens, (k=1,,8), redues he ompuaion omplexiy and also derease he image qualiy. And large quanizaion sep size will be used. As he low frequeny oeffiiens monopolize mos of he signal energy, here will be a lile informaion loss. While here is obje moving or inruding in he monioring sene, he monioring image sequenes are of grea ineres o he user. To ahieve he disorion requiremen, all he CT oeffiiens will be proessed, and small quanizaion sep size will be used. The required number of mulipliaions ( M ) and he oal number of addiions ( A ) for he N N CT is given by following equaions, N is a power of wo. M ( N / )log N (7) A N N N (8) (5 / )log runing algorihm alulaing lower CT oeffiiens, N0 is a power of wo, he number of mulipliaions ( M N ) and addiions ( A 0 N0 N0 ) for he ompuaion of N0 N0 poins are M N N (9) N0 (3 / 4) log 0 A ( N N )log N N 3N N (10) The quanizaion able is enral o he ompression/fideliy performane of JEG oding. For he image ompression i is neessary o develop a new quanizaion able o ahieve a beer energy saving han he JEG sandard quanizaion able. V. SIMULATION RESULTS 130

12 SNR Xiong Zheyuan, Fan Xiaping, Liu Shaoqiang, Li Yongzhou and Zhang Huan, erformane Analysis For CT-Based Coded Image Communiaion in Wireless Mulimedia Sensor Neworks To analyze he relaionship beween enoding omplexiy, image disorion and image daa rae, several simulaions are ondued using Malab. Figure shows how he SNR evolves wih k for he hree images. I is obviously ha he energy onsumpion for image ompression is inreasing as he k inreases, herefore he SNR inreases. The es image sequenes Lena, Window and Beah are in differen resoluions. Lena represens he images ha onain muh informaion, whih an be lassified as a high frequeny images wih high spaial deails. Window represens he images ha onain lile informaion and an be lassified as a high frequeny image wih low spaial onen. Beah represens images ha onaining less informaion and an be lassified as low frequeny image wih low spaial deails. runing algorihm alulaing k k par of he CT oeffiiens, (k=1,,8), redues he ompuaion omplexiy and also derease he image qualiy. Figure show how he SNR evolves wih k for he hree images. I is obviously ha he number of mulipliaions, addiions and memory aesses for CT is inreasing as he k inreases, herefore he energy onsumpion for image ompression inreases. The fideliy loss in JEG oding ours enirely in quanizaion, and muh of he ompression is gained by run lengh oding of oeffiiens wih quanize o zero. Therefore he quanizaion able is enral o he ompression/fideliy performane of JEG oding. For he image ompression i is neessary o develop a new quanizaion able o ahieve a beer energy saving han he JEG sandard quanizaion able. 35 Lena Window Beah k 131

13 SNR SNR INTERNATIONAL JOURNAL ON SMART SENSING AN INTELLIGENT SYSTEMS VOL. 6, NO. 1, FEBRUARY 013 Figure. SNR variaions wih k for hree images Quanizaion 1 JEG sandard Quanizaion Bi per pixel Figure 3. SNR omparisons of hree quanizaion ables on Lena image Quanizaion 1 JEG sandard Quanizaion Bi per pixel Figure 4. SNR omparisons of hree quanizaion ables on eppers image JEG s quanizaion sep sizes largely deermine he rae-disorion radeoff in a JEG ompressed image. Using he JEG sandard quanizaion ables is subopimal sine hey are 13

14 Xiong Zheyuan, Fan Xiaping, Liu Shaoqiang, Li Yongzhou and Zhang Huan, erformane Analysis For CT-Based Coded Image Communiaion in Wireless Mulimedia Sensor Neworks image-independen. Therefore, i is opimal o adap he quanizaion able aording o he sene aiviy in WMSN. Comparaive performane is sudied in erms of SNR onsidering hree quanizaion ables. Quanizaion able used in JEG algorihm is applied o quanize he oeffiiens denoed as JEG sandard. Quanizaion able used in sai sene image frame denoed as Quanizaion 1. Quanizaion able used in moion sene image frame denoed as Quanizaion. Quanizaion simulaion resuls for image Lena and eppers are shown in Figure 3 and Figure 4. We es he proposed image ompression sheme on ETS 000 (IEEE Inernaional Workshop on erformane Evaluaion of Traking and Surveillane 000) es daases, whih are he oudoor parking lo surveillane images of people and vehile raking. Tes image sequene is onsised of 145 frames a 5 Hz (58.08 seonds), single amera. The original image is JEG forma olor image,whih is swihed ino he ommonly used surveillane image forma SIF (soure inpu forma) ha is gray sale image. Table1 shows how he ompression raio affes he image qualiy of JEG ompression and frame rae for differen values of he qualiy of JEG ompression. I is obviously ha if ompression raio inreases, he image qualiy inreases, bu frame rae dereases. Table 1 Image qualiy and Maximum number of images per seond ha an be ransmied for differen ompression raio Compression rae Image qualiy Frame rae Q0 1.9%.17 Q1 1.94% Q.80% Q3 3.61% 7.89 Q4 4.30% 6.6 Q5 5.01% 5.69 Q6 5.78% 4.94 Q7 7.01% 4.07 Q8 9.08%

15 INTERNATIONAL JOURNAL ON SMART SENSING AN INTELLIGENT SYSTEMS VOL. 6, NO. 1, FEBRUARY 013 Q %.0 Q % 0.6 VI. CONCLUSION In his paper, based on he sudy of CT based oded image ommuniaion performane on video sensors in WMSN, energy onsumpion and image disorion models for image ompression and ransmission under energy onsrains are derived. Then an energy-minimized resoure alloaion sheme is presened, whih adap enoder omplexiy aording o he video sene aiviy. Simulaion resuls indiae ha our proposed sheme an save sensor energy signifianly exep for small qualiy degeneraion of reonsrued images and a lile more oupied ransmission raes. REFRENCES [1] Akyildiz I F, Melodia T, Chowdhury K R. Wireless mulimedia sensor nework: a survey, IEEE Transaions on Wireless Communiaion, Vol.14, no.6, 007, pp [] Muhammad F. Sabir, Hamid Rahim Sheikh, Rober W. Heah, Jr., and Alan C. Bovik. A Join Soure-Channel isorion Model for JEG Compressed Images, IEEE Transaions on Image roessing, Vol.15, no.6, 006, pp [3]So udlewski, Tommaso Melodia. A disorion-minimizing rae onroller for wireless mulimedia sensor neworks, Compuer Communiaions, Vol.33, 010, pp [4] Zhihai He, Wenye Cheng, and Xi Chen. Energy minimizaion of porable video ommuniaion devies based on power-rae-disorion opimizaion, IEEE Transaions on Cirui and Sysems for Video Tehnology, Vol.18, no.5, 008, pp [5] Zhao Sun, Xi Chen, and Zhihai He. Adapive Crii esign for Energy Minimizaion of orable Video Communiaion evies, IEEE Transaions on Ciruis and Sysems for Video Tehnology, Vol.0, no.1, 010, pp [6] Zhenhua Tang, Tuanfa Qin, and Wenyu Liu. Energy-Minimized Adapive Resoure Alloaion for Image Transmission over Wireless Channel, roeedings of Inernaional Conferene on Inelligen Conrol and Informaion roessing, IEEE, 010, pp

16 Xiong Zheyuan, Fan Xiaping, Liu Shaoqiang, Li Yongzhou and Zhang Huan, erformane Analysis For CT-Based Coded Image Communiaion in Wireless Mulimedia Sensor Neworks [7] Mammeri A, Khoumsi A, Ziou, e al. Energy-aware JEG for visual sensor neworks, roeedings of The 008 Maghrebian Conferene on Sofware Engineering and Arifiial Inelligene, IEEE, 008, pp [8] Mammeri A, Khoumsi A, Ziou, e al. Modeling and adaping JEG o he energy requiremens of visual sensor neworks, roeedings of 008 Inernaional IEEE Workshop on Sensor Neworks, 008, pp [9] Mammeri A, Khoumsi A, Ziou, e al. Energy effiien ransmission sheme of JEG images over VSN, roeedings of 008 Inernaional IEEE Workshop on erformane and Managemen of Wireless and Mobile Neworks, 008, pp [10] Z. He, J. Cai, and C. W. Chen. Join soure hannel rae-disorion analysis for adapive mode seleion and rae onrol in wireless video oding, IEEE Transaion on Cirui and Sysems for Video Tehnology, Vol.1, no.6, 00, pp [11] Z. He, Y. Liang, L. Chen, I. Ahmad, and. Wu. ower-rae-disorion analysis for wireless video ommuniaion under energy onsrain, IEEE Transaion on Cirui and Sysems for Video Tehnology, Vol.15, no.5, 005, pp [1] Z. He and. Wu. Resoure alloaion and performane limi analysis of wireless video sensors, IEEE Transaion on Cirui and Sysems for Video Tehnology, Vol.16, no.5, 006, pp [13] Lee, Kim H, Tu S, e al. Energy-effiien image ompression on resoure-onsrained plaforms, IEEE Transaion on Image roessing, Vol.18, no.9, 009, pp