human behavior understand behavior properti spatio dtempor local featur visual transform sonifi face facial express sound gener dconvent method content analysi understand panel overview pervas video analysi workshop overview human dcenter multimedia system tutori overview automat gener soundtrack low dlevel cross dmodal correl automot scenario put piec multimod analysi social attent meet ey fixat databas salienc detect imag challeng human behavior understand abstract recent advanc pattern recognit allow comput scientist psychologist jointli address automat analysi human behavior comput workshop human behavior understand intern confer pattern recognit explor number aspect open question field demonstr multi disciplinari natur research area summari give overview workshop discuss main research challeng emploi social gaze speak activ automat determin extravers trait visual gaze estim joint head ey inform paper present unconstrain visual gaze estim system propos method extract visual field view person target scene order estim approxim locat interest visual gaze novelti system joint head pose ey locat inform fine tune visual gaze estim head pose system multipl scenario improv obtain propos approach valid boston univers head pose dataset standard deviat joint visual gaze estim improv horizont vertic respect gaze estim obtain head pose user studi show potenti propos system automat face gestur recognit automat face gestur recognit larg dscale imag video search challeng technolog trend exploit facial express affect video summaris webcam dbase visual gaze estim abstract paper combin state art ey center locat ey corner locat system estim visual gaze user control environ sit front screen order reduc minimum comput cost ey corner locat built technolog ey center locat tweak specif task high map precis prioriti applic claim system achiev accept accuraci requir addit dedic hardwar bring gaze base methodolog human comput interact mainstream isocentr color salienc imag human dcenter comput applic multimedia proceed acm workshop human dcenter comput intern workshop human dcenter comput hcc facial express recognit creativ interfac present audiovisu creativ tool automat recogn facial express real time produc sound combin imag facial express recognit compon detect track face output featur vector motion specif locat face featur vector input bayesian network classifi facial express categori angri disgust happi classif result featur vector gener combin sound imag chang real time depend person facial express explain basic compon tool applic art perform instal medic domain similar match comput vision multimedia compar imag imag model fundament oper retriev system similar match imag resid hierarch level pixel pixel level featur space level object level semant level system interest simpl pixel pixel comparison suffic differ determin bear correl perceptu differ imag differ adequ semant imag similar match techniqu develop recognit object condit distort similar measur hand applic imag databas match dissimilar measur seldom base techniqu differ emphasi applic distanc learn similar estim abstract paper present gener guidelin find distanc measur similar estim base statist analysi distribut model distanc function set distanc measur deriv harmon distanc geometr distanc gener variant maximum likelihood theori measur provid accur featur model classic euclidean manhattan distanc find featur element heterogen sourc influenc similar estim assumpt singl isotrop distribut model inappropri allevi problem boost distanc measur framework find multipl distanc measur fit distribut select featur element accur similar estim distanc measur similar estim test applic stereo match motion track video sequenc perform boost distanc measur evalu benchmark data set uci repositori imag retriev applic experi robust result obtain base propos method special section acm multimedia confer proceed intern confer multimedia proceed acm intern confer imag video retriev proceed acm intern confer imag multimedia content analysi mine www infotech monash human dcomput interact www bc hci org uk find happen hci interact music aeroplan petrol pump publish british hci group issn human dcenter multimedia system tutori overview overview imageclef object retriev task abstract describ object retriev task imageclef give overview method particip group present discuss result task base wide pascal object recognit data train object recognit method iapr tc benchmark dataset imag object ten class bicycl buse car motorbik cat cow dog hors sheep person retriev intern group particip wide varieti method result evalu show task challeng method relev assess strong influenc result evalu integr relev feedback boost content dbase imag retriev ddimension adapt discrimin analysi human dcomput intellig interact survei human dcenter comput challeng perspect cooper object track multipl ptz camera colour interest point improv imag retriev imag retriev scenario method interest point detect earli stage find region descriptor calcul find salient locat imag data crucial task observ current method lumin inform imag investig colour inform interest point detect multi channel inform harri corner detector explor colour space evalu determin characterist scale interest point colour scale select method present show colour inform boost salient colour result improv perform retriev task person multimedia retriev trend aim paper present overview current situat hot topic multimedia inform retriev person aspect problem hand user person access imag video collect achiev provid intuit natur brows capabl custom featur system requir perform user profil adapt exist paramet system user interest hand import devic applic technolog deploi mobil media high growth area state art technolog lag consum expect address paper precis import aspect multimod human dcomput interact survei abstract paper review major approach multimod human comput interact give overview field comput vision perspect focu bodi gestur gaze affect interact facial express recognit emot audio discuss user task model multimod fusion highlight challeng open issu emerg applic multimod human comput interact mmhci research guest editor introduct human dcenter comput dtoward human revolut context dbase object dclass recognit retriev gener correlogram abstract present approach retriev object categori base type imag represent gener correlogram gc imag represent object constel gc encod inform local part spatial relat part part context show represent fast procedur learn object categori weak supervis effici match model object larg collect imag learn stage show integr represent boost system obtain compact model repres featur featur convei kei properti object part spatial arrang match step propos direct procedur exploit represent effici spatial coher match local part combin data organ invert file show thousand imag evalu effici framework appli standard databas show result favor compar state art method comput cost accuraci index term object recognit retriev boost spatial pattern contextu inform ag human comput interact paper present set evid base guidelin design websit elderli extens literatur review conduct guidelin develop support literatur final guidelin evalu ten websit primari target audienc elderli problem address lot past research topic ag human comput interact elderli central collect inform present organis studi structur classif usabl guidelin creat address area human comput interact elderli inform research web design convei inform websit structur inform attract visitor attent common find issu address call typic audienc mind frequent user world wide web frequent user world wide web young gener default world wide web profit busi capitalis main asset inform peopl visit websit inform retriev purpos defeat peopl visit websit unabl obtain requir inform inform hidden websit design techniqu dispos attract attent content websit colour imag anim techniqu logic plausibl issu websit access peopl includ young gener disabl elderli find guidelin make experi inform retriev websit userfriendli gener feasibl gener guidelin make websit access elderli specif traceabl elderli import authent facial express analysi argu comput interact human commun skill human skill abil understand emot state person express human displai emot facial express facial express system databas emot data collect subject perform seri facial express direct deliber facial action task typic differ appear time authent facial express induc event normal environ subject paper present effort creat authent facial express databas base spontan emot deriv environ test compar wide rang classifi machin learn literatur facial express classif comput vision human dcomput interact present nonintrus system base comput vision human comput interact dimension environ control hand point gestur user allow walk room manipul inform displai wall hand point devic captur track real time stereo vision hand point gestur remap current point interest reproduc advanc interact scenario drag click behavior tradit mice system call pointat patent pend enjoi care model user optic subsystem visual algorithm calibr adapt user peculiar environment conclud section provid insight system characterist perform relev real applic human dcenter comput multimedia perspect corner detector affin invari salient region color import robust distanc metric analysi similar estim paper present gener guidelin establish relat distribut model similar estim rich set distanc metric harmon distanc geometr distanc deriv maximum likelihood theori metric provid accur featur model convent euclidean distanc ssd manhattan distanc sad featur element heterogen sourc influenc similar estim assumpt singl isotrop distribut model inappropri propos boost distanc metric find distanc metric fit distribut underli element select import featur element respect similar experi distanc metric similar estim comput studi distanc metric similar measur distanc metric wide similar estim paper find popular euclidean manhattan distanc suitabl data distribut gener guidelin establish relat distribut model similar estim propos base maximum likelihood theori propos distanc metric harmon distanc geometr distanc featur element heterogen sourc influenc similar estim inappropri model distribut isotrop propos boost distanc metric find distanc metric fit distribut underli element select import featur element respect similar boost distanc metric test fifteen benchmark data set uci repositori imag retriev applic experi robust result obtain base propos method emot recognit base joint visual audio cue role featur configur inform face classif simul expertis hypothesi boost distanc estim applic dnearest neighbor classifi content dbase multimedia inform retriev state art challeng extend boundari scienc art cultur content base multimedia inform retriev paradigm method search myriad varieti media world survei review recent articl content base multimedia inform retriev discuss role current research direct includ brows search paradigm user studi affect comput learn semant queri featur media type high perform index evalu techniqu base current state art discuss major challeng futur comput vision human dcomput interact present nonintrus system base comput vision human comput interact dimension environ control hand point gestur user allow walk room manipul inform displai wall hand point devic captur track real time stereo vision hand point gestur remap current point interest reproduc advanc interact scenario drag click behavior tradit mice system call pointat patent pend enjoi care model user optic subsystem visual algorithm calibr adapt user peculiar environment conclud section provid insight system characterist perform relev real applic affect multimod human dcomput interact paper describ intellig system develop support affect multimod human comput interact amm hci user action emot model adapt hci support user activ propos system name gaze base sens interpret human part comput context integr number natur human commun modal includ speech ey gaze direct face facial express number standard hci modal keystrok mous movement activ softwar identif turn fed process provid decis make adapt hci support user activ prefer attain system educ improv knowledg decis make experi case base reason infer engin gaze util case base dynam increment organ event content address memori fact retriev evalu encount event base user prefer gener form prior input support concept concurr modular scalabl persist mobil gaze built agent base system agent respons part process usabl studi conduct offic scenario number user gaze perceiv effect easi affect qualit fast spatial pattern discoveri integr boost constel contextu descriptor present approach fast object class recognit incorpor contextu inform boost object repres constel gener correlogram integr inform local part spatial relat incorpor spatial relat constel descriptor show exhaust search match avoid combin contextu descriptor boost system simultan learn inform character part object characterist mutual spatial relat propos framework includ match step homolog part train set learn spatial pattern match match part approach provid supervis algorithm unsupervis result favor compar state art result multimod human comput interact survei abstract paper review major approach multimod human comput interact give overview field comput vision perspect focu bodi gestur gaze affect interact facial express recognit emot audio discuss user task model multimod fusion highlight challeng open issu emerg applic multimod human comput interact mmhci research neighborhood issu singl dframe imag super dresolut video object boundari reconstruct dpass vote affect meet video analysi paper examin affect content meet video ask subject manual label meet video continu respons measur continu scale label real time energi valenc dimens human affect space automat extract audio visual featur character affect content video compar result manual label low level automat audiovisu featur extract analysi yield promis result suggest affect meet video analysi lead interest observ automat index effici object dclass recognit boost contextu inform complet perform graph content dbase imag retriev add gener normal scope system architectur multimedia inform retriev learn probabilist classifi human dcomput interact applic proceed acm sigmm intern workshop multimedia inform retriev comput vision human dcomput interact present nonintrus system base comput vision human comput interact dimension environ control hand point gestur user allow walk room manipul inform displai wall hand point devic captur track real time stereo vision hand point gestur remap current point interest reproduc advanc interact scenario drag click behavior tradit mice system call pointat patent pend enjoi care model user optic subsystem visual algorithm calibr adapt user peculiar environment conclud section provid insight system characterist perform relev real applic state dof dthe dart human dcomput interact unic establish framework multimod languag dialog framework evolv spoken exchang import aspect develop human center inform system import issu achiev synerg man machin term human center emphas fact exist inform system design human user mind user friendli scientif engin commun effect chang inform system ubiquit human endeavor includ scientif medic militari transport consum individu user learn search inform includ data mine research includ visual comput author multipl user group user group group user commun collabor singl multipl authent emot detect real dtime video grow trend emot intellig humancomput interact paradigm order react appropri human comput percept emot state human authent facial express analysi argu comput interact human commun skill human skill abil understand emot state person express human displai emot facial express facial express system databas emot data collect subject perform seri facial express direct deliber facial action task typic differ appear time authent facial express induc event normal environ subject paper present effort creat authent facial express databas base spontan emot deriv environ test compar wide rang classifi machin learn literatur facial express classif improv error metric comput vision algorithm euclidean ssd sum squar differ metric preval justifi maximum likelihood perspect addit nois gaussian gaussian nois distribut assumpt invalid previou research found metric doubl exponenti metric cauchi metric provid result accord maximum likelihood approach paper examin error metric provid theoret approach deriv rich set nonlinear estim result imag databas show robust result obtain nois estim base propos error metric analysi analysi unlabel data semi dsupervis learn imag retriev increas interest unlabel data semi supervis learn classif problem previou work show unlabel data improv degrad classif perform depend model assumpt match ground truth data distribut complex classifi compar size label train set paper provid analysi unlabel data distribut label unlabel data show migrat effect semi supervis learn extens experi perform context imag retriev applic approach evalu unlabel data aspect aim provid guidelin unlabel data skin detect bayesian network approach autom detect track human comput vision necessit improv model human skin appear paper propos bayesian network approach skin detect test classifi propos methodolog incorpor unlabel data appli semi supervis approach skin detect show learn structur bayesian network classifi enabl learn good classifi small label set larg unlabel set robust error metric analysi nois estim imag index boost contextu inform content dbase imag retriev complet perform graph probabilist inform retriev authent emot recognit semisupervis learn classifi theori automat classif basic task requir pattern recognit human comput interact applic paper discuss train probabilist classifi label unlabel data provid analysi show condit unlabel data learn improv classif perform show condit violat unlabel data detriment classif perform discuss implic analysi specif type probabilist classifi bayesian network propos structur learn algorithm util unlabel data improv classif final show result algorithm successfulli emploi applic relat human comput interact pattern recognit facial express recognit face detect proceed acm sigmm intern workshop multimedia inform retriev imag video retriev state art imag video retriev evalu express recognit techniqu express human displai emot facial express work report advanc made build system classif facial express continu video input introduc test di erent bayesian network classifi classifi express video naiv bay classifi learn depend di erent facial motion featur tree augment naiv bay tan classifi investig neural network approach propos architectur hidden markov model hmm automat segment recogn human facial express video sequenc explor person depend person independ recognit express compar di erent method learn bayesian network classifi facial express recognit label unlabel data understand human emot skill comput interact intellig human user express human displai emot facial express paper report advanc made build system classif facial express continu video input bayesian network classifi classifi express video motiv factor bayesian network classifi abil handl miss data infer train interest problem learn label unlabel data show unlabel data learn classifi correct model assumpt critic achiev improv classif perform motiv introduc classif driven stochast structur search algorithm learn structur bayesian network classifi show moder size label train set larg amount unlabel data method util unlabel data improv classif perform provid result naiv bay nb tree augment naiv bay tan classifi show achiev good perform label train set perform poorli unlabel data ad train set content dbase index perform size normal precis semi dsupervis learn facial express recognit automat classif machin basic task requir pattern recognit human comput interact applic paper discuss train probabilist classifi label unlabel data provid analysi show condit unlabel data learn improv classif perform discuss implic analysi specif type probabilist classifi bayesian network propos structur learn algorithm util unlabel data improv classif final show result algorithm successfulli emploi facial express recognit applic facial express recognit video sequenc tempor static model human comput intellig interact hcii emerg field scienc aim provid natur wai human comput aid argu comput interact human commun skill human skill abil understand emot state person express human displai emot facial express work report advanc made build system classif facial express continu video input introduc test architectur focus distribut assumpt featur depend structur introduc facial express recognit live video input tempor cue method tempor inform extens explor speech recognit applic method templat match dynam program method hidden markov model hmm work exploit exist method propos architectur hmm automat segment recogn human facial express video sequenc architectur perform segment recognit facial express automat multi level architectur compos hmm layer markov model layer explor person depend person independ recognit express compar method video retriev summar present system retriev summar continu archiv multimedia data home ubiquit environ data pressur base floor sensor analyz index video audio larg number sourc video audio handov implement retriev continu video stream sound person move environ kei frame extract propos algorithm implement obtain compact summari activ person cluster algorithm imag analysi identifi action event system consist graphic user interfac retriev video summari interact simpl queri evalu salient point techniqu imag retriev global featur relat color textur commonli describ imag content probleln approach global featur captur part imag characterist local comput imag inform salient point repres local inform discrimin featur comput paper compar wavelet base salient point extract algorithm corner detector criteria repeat rate inform content show extract color textur inform locat onr salient point significantli improv result term retriev accuraci comput complex storag space featur vector compar global featur approach compar salient point detector of point incont bash retrievalallow imag index torepresq localproperti imag clase cornerdetector al beusz forthi purpo havedrawback appli tovari natur imag becau viss featur becorn andcorn gather in qh region inthi paper wepreshh point detectorusec wavelettranst compar cornerdetector usec criteria repeat rate inform content determin detectorg besresqh sd st criteria imag video retriev challeng imag video retriev robust shape match visual match algorithm term featur inter featur distanc metric commonli metric sum squar di erenc ssd valid maximum likelihood perspect real nois distribut gaussian found experiment gaussian nois distribut assumpt invalid impli metric distribut closer real nois distribut paper consid shape match applic object recognit video retriev recognit object video offer signific benefit video retriev includ automat annot content base queri base object characterist paper describ preliminari work recogn object video sequenc survei relev research literatur kalman filter obtain segment blob video classifi blob probabl ratio test appli tempor filter method result sequenti classif method video sequenc blob result real video sequenc shown evalu salient point techniqu imag retriev global featur relat color textur commonli describ imag content probleln approach global featur captur part imag characterist local comput imag inform salient point repres local inform discrimin featur comput paper compar wavelet base salient point extract algorithm corner detector criteria repeat rate inform content show extract color textur inform locat onr salient point significantli improv result term retriev accuraci comput complex storag space featur vector compar global featur approach emot recognit cauchi naiv bay classifi recogn human facial express emot comput interest challeng problem paper propos method recogn emot facial express displai video sequenc introduc cauchi naiv bay classifi cauchi distribut model distribut provid framework choos model distribut assumpt person depend person independ experi show cauchi distribut assumpt typic result gaussian distribut assumpt detect automobil peopl semant video retriev paper describ method detect automobil peopl stream archiv video video object track system base kalman filter updat activ contour video sequenc sequenti probabl ratio test sprt classifi move object result shown real video sequenc busi citi intersect salient point content dbase retriev imag retriev global featur relat color textur commonli describ imag content problem approach global featur captur part imag characterist local comput imag inform salient point repres local inform discrimin featur comput show extract color textur inform locat salient point significantli improv result compar global featur approach salient point lead complet imag represent corner extend perform graph cluster retriev perform evalu probabilist inform retriev present precis recal precisionscop graph avoid domin effect embed irrelev fraction precis recal valu offer incomplet overview inform retriev system studi inform system paramet gener embed relev fraction random perform effect vari scope badli miss compar salient point detector of point incont bash retrievalallow imag index torepresq localproperti imag clase cornerdetector al beusz forthi purpo havedrawback appli tovari natur imag becau viss featur becorn andcorn gather in qh region inthi paper wepreshh point detectorusec wavelettranst compar cornerdetector usec criteria repeat rate inform content determin detectorg besresqh sd st criteria content dbase imag retriev wavelet dbase salient point content base imag retriev cbir activ research area past year attent research focus index techniqu base global featur distribut global distribut limit discrimin power unabl captur local imag inform appli global gabor textur featur greatli improv retriev accuraci computation complex paper present wavelet base salient point extract algorithm show extract color textur inform locat point significantli improv result term retriev accuraci comput complex storag space featur vector compar global featur approach imag retriev wavelet dbase salient point content base imag retriev cbir activ research area past year attent research focus index techniqu base global featur distribut global distribut limit discrimin power unabl captur local imag inform appli global gabor textur featur greatli improv retriev accuraci computation complex paper present wavelet base salient point extract algorithm show extract color textur inform locat point significantli improv result term retriev accuraci comput complex storag space featur vector compar global featur approach color dbase retriev content base retriev color index preval retriev method kei problem color index determin color space find distanc measur attent research literatur focuss color model consider nois model focus nois model show find distanc measur optim sens maxim similar probabl result includ experi color base retriev object recognit implement test promis algorithm research literatur benchmark introduct world enter digit ag visual media preval easili access factor explos growth world wide web terabyt disk server digit versatil disk reveal grow amount visual media societi maximum likelihood stereo match research literatur maximum likelihood principl appli stereo match alter stereo pair differ gaussian distribut paper present method appli maximum likelihood stereo match approach measur real nois distribut train set construct metric denot maximum likelihood metric compar stereo pair maximum likelihood metric optim sens maxim probabl similar experi discuss compar maximum likelihood metric promis algorithm research literatur intern stereo data set show algorithm research literatur improv maximum likelihood metric sum squar differ introduct common assumpt real nois distribut fit eith improv visual match visual match algorithm term featur inter featur distanc metric commonli metric sum squar differ ssd valid maximum likelihood perspect real nois distribut gaussian base real nois distribut measur intern test set found experiment gaussian nois distribut assumpt invalid impli metric distribut closer real nois distribut paper consid visual match applic content base retriev imag databas stereo match broaden result implement sophist algorithm research literatur algorithm compar efficaci ssd metric sad sum absolut differ metric cauchi metric kullback rel inform visual websearch icon queri imagin profession web site design constantli innov client homepag find content major search engin hotbot http www hotbot find text web typic capabl find visual media articl discuss method find visual media www emphasi icon queri essenti drag drop visual concept simpl semant describ method find static featur set describ method call activ featur set choos featur set base context imag wavelet dbase salient point imag retriev interest point content base imag retriev imag index repres local properti imag classic corner detector purpos drawback appli natur imag imag retriev visual featur corner corner gather small region maximum likelihood stereo match research literatur maximum likelihood principl appli stereo match alter stereo pair differ gaussian distribut paper present method appli maximum likelihood stereo match approach measur real nois distribut train set construct metric denot maximum likelihood metric compar stereo pair maximum likelihood metric optim sens maxim probabl similar experi discuss compar maximum likelihood metric promis algorithm research literatur intern stereo data set show algorithm research literatur improv maximum likelihood metric sum squar differ introduct common assumpt real nois distribut fit eith wavelet base textur classif textur basic featur visual search comput vision literatur attent focuss textur featur minim consider nois model paper investig problem textur classif maximum likelihood perspect account textur model nois distribut inter depend textur featur investig show real nois distribut closer exponenti gaussian distribut metric retriev rate propos cauchi metric altern metric provid direct method deriv optim distort measur real nois distribut experiment consist improv result metric conclud result discuss intern textur databas introduct color base retriev recognit content base retriev color index preval retriev method kei problem color index determin color space find distanc measur attent research literatur focuss color model consider nois model focus nois model show find distanc measur optim sens maxim similar probabl result includ experi color base retriev object recognit implement test promis algorithm research literatur benchmark introduct world enter digit ag visual media preval easili access factor explos growth world wide web terabyt disk server digit versatil disk reveal grow amount visual media societi wavelet dbase salient point applic imag retriev color textur featur imag retriev global featur relat color textur commonli describ imag interest point contentbas imag retriev imag index repres local properti imag ground dtruth train set hierarch cluster content dbase imag retriev progress content base imag retriev cbir hamper absenc document valid test set provid ground truth perform evalu imag index retriev cluster task quick access larg tenthousand million imag digit imag collect hierarch structur index brows mechan base cluster similar imag coars fine level highli want leiden centuri portrait databas lcpd consist scan studio portrait call cart de visit cdv delin set cluster studio logo backsid imag cluster similar semant ident logo form number level show clear hierarchi leiden imag multimedia group construct cd rom document set studio portrait logo serv ground truth featur perform evalu domain color index grei level imag lai characterist precalcul featur vector set portrait copi pair studio logo cluster ident logo test set provid cluster level statist signific number test set imag embed realist larg environ narrow domain imag present cbir commun enabl select optim index retriev approach part internation defin test set compris test set specif design color textur shape retriev evalu improv rank metric abstract comput vision algorithm metric similar measur determin distanc featur euclidean ssd sum squar differ metric preval justifi maximum likelihood perspect addit nois distribut gaussian base real nois distribut measur intern test set found gaussian nois distribut assumpt invalid impli metric distribut closer real nois distribut paper applic content base retriev imag databas stereo match motion track experi model function nois distribut comput accuraci method distanc measur experi compar ssd metric sad sum absolut differ metric cauchi metric kullback rel inform algorithm research literatur ssd sad show greater accuraci obtain cauchi metric index term maximum likelihood rank metric content base retriev color index stereo match motion track robust color index content base imag retriev color index preval retriev method literatur attent focuss color model consider nois model paper investig problem color index maximum likelihood perspect account color model nois distribut quantiz color featur real nois distribut deriv distort measur consist improv accuraci investig conclud result real stock photographi databas consist color imag introduct visual media retriev method color index domin method shown effect academ commerci arena color index histogram method feasibl term memori usag provid suffici accuraci histo multi dscale dimag search imag retriev subregion larg imag databas immens number queri index encod spatial inform imag make assumpt queri addit index scale object imag paper propos approach hierarch index encod imag region gain fix partit suggest index color featur easi implement index test databas imag introduct imag retriev system retriev imag base imag comparison user typic interest object search interest subregion interest object imag queri system retriev imag subregion human percept databas task call imag adapt dd tree visual retriev frequent occur problem imag retriev find similar imag gener find nearest neighbor literatur tree effici method find nearest neighbor high dimension space paper survei relev tree literatur adapt promis solut problem imag retriev find paramet bucket size threshold test system corel studio photo databas imag measur user respons time retriev accuraci introduct search multimedia outstand problem world wide web compani network databas philip media tera byte databas deriv ownership trademark video imag audio real world usag multimedia search program support enabl trade asset gener hdl da dmodel micromembran finit delement ddescript 