简介:Objectivevideoqualityassessmentmethodsoftenevaluatealltheframesregardlessoftheirimportance.Forwirelessdistortedvideos,noteveryframehasthesamecontributiontothefinaloverallqualityduetothechannelfadingandinterference,whichmayleadtothecapacityvariationintemporal.Besides,withthecontentsimilarityanderrorpropagationpatternintemporaldomain,itispossibletoevaluatetheoverallqualitywithonlypartoftheframes.Inthispaper,ademonstrationisperformedtoshowthatthevideoqualitycanbeevaluatedwithreducedframesset(RFS),andastatetransitionmodelisproposedtoextracttheRFS.Atlast,avideoqualityassessment(VQA)methodiscarriedoutbasedonRFS.Comparedwithseveralstate-of-the-artmethods,ourmethodcanachieveasuitableaccuracywithlessframestobeprocessed.
简介:聪明的边网络内容条款热点的出现,与巨大的存储空间被装备,,几GB),开创机会学习在边网络交付录像的可能性。与两个不同常规内容条款交货网络(CDN)并且对等(P2P)策划,这个新交货范例,也就是边录像CDN,要求在设法并列地要服务的用户家/办公室定位的多达百万个边热点活动录像内容条款。明确地,二挑战涉及造边录像CDN,包括边内容条款热点怎么应该被组织为用户服务,并且内容项目怎么应该在为用户服务的不同地点被复制到他们。探讨这些挑战,我们如下建议我们的数据驱动的设计。首先,我们提出一个边区域分区问题联合最大化用户经验丰富的质量并且最小化复制费用,它在自然是NP难的,并且我们设计一个象Voronoi一样分区算法产生最佳的服务房间。第二,满足条款热点复制内容项目到边网络,我们建议边请求预言基于复制策略,它在卸载举止的一座服务者山峰执行复制。我们实现我们的设计并且使用踪迹驱动的实验验证它的有效性。与常规集中的CDN和基于流行的复制相比,我们的设计能显著地改进经验的用户质量,以用户察觉了带宽和潜伏,直到40%。
简介:与WiFi和3G/4G的快速的发展,人们趋于在移动设备上看录像。这些设备是无所不在的,但是有小存储器缓冲录像。作为结果,与传统的计算机相对照,这些设备加重内容供应商的网络压力。以前的研究使用CDN解决这个问题。但是它出租空间不能动态地在被调整的静态的租借机制让运作的费用高飞并且与不兼容动态地录像交货。在我们的学习,基于从Tencent录像的用户行为的彻底的分析,一个流行中国联机录像份额平台,我们识别二关键用户行为。第一,在一样的区域的大量用户趋于看一样的录像。第二,录像的流行分发符合Pareto原则,即,20%流行录像拥有的顶80%所有录像交通。把这些观察变成银子弹,我们在需求系统(CPA-VoD)上建议并且实现一个新奇帮助云、帮助同伴的录像。在系统,我们在象同伴群的一样的区域,并且在一样的同伴群组织用户,用户们能由分享他们的缓冲录像提供录像给另外的用户。而且,我们在云服务器缓冲10%很流行的录像进一步减轻网络压力。我们选择云服务者因为出租空间能动态地被调整,缓冲录像。根据从Tencent录像的真实数据集上的评估,CPA-VoD最优地减轻网络压力和操作费用,当仅仅20.9%交通被内容供应商满足时。
简介:Inthispaper,weproposeasusceptible-infected-susceptible(SIS)modeloncomplexnetworks,small-world(WS)networksandscale-free(SF)networks,tostudytheepidemicspreadingbehaviorwithtimedelaywhichisaddedintotheinfectedphase.Consideringtheuniformdelay,thebasicreproductionnumberR_0onWSnetworksandR_0onSFnetworksareobtainedrespectively.OnWSnetworks,ifR_0≤1,thereisadisease-freeequilibriumanditislocallyasymptoticallystable;ifR_0>1,thereisanepidemicequilibriumanditislocallyasymptoticallystable.OnSFnetworks,ifR_0≤1,thereisadisease-freeequilibrium;ifR_0>1,thereisanepidemicequilibrium.Finally,wecarryoutsimulationstoverifytheconclusionsandanalyzetheeffectofthetimedelayr,theeffectiverateλ,averageconnectivityandtheminimumconnectivitymontheepidemicspreading.
简介:Renderingoflarge-scaleforestscenesisachallengingtask,whosehighlygeometriccomplexitywillputheavyburdenoncurrentgraphicshardware.Whennavigatingthescene,theoverallvisualresultisgenerallyconsideredasthecoreconcern.Anewmethodisproposedinthispaperforlarge-scaleforestrenderingusingclusteringandmergingstrategies.Ourmethodimprovestherenderingeffectbyclusteringpolygonsaccordingtothepointinformationwithrelationtoneighbours.Afastforestrenderingsystemisdevelopedaccordingly.Therelativetechniquesinthesystemcanimprovethevisualqualityondemandofdifferentapplications.
简介:Thenoisedatainverticalcomponentrecordsof85seismicstationsinFujianProvinceduring2012isusedastheresearchobjectinthispaper.Thenoisedataisdividedintofiveminutesegmentstocalculatethepowerspectra.Thehighreferencelineandlowreferencelineofstationarethenidentifiedbydrawingaprobabilitydensityfunctiongraph(PDF)usingthepowerspectralprobabilitydensityfunction.Moreover,accordingtotheanomaliesofPDFgraphsin85seismicstations,theabnormalnoiseisdividedintofourcategories:droppedpacket,lownoise,highnoise,andmediannoiseanomalies.Afterwards,fourselectionmethodsarefoundbythehighorlownoisereferencelineofthestations,andthesystemofreal-timemonitoringofseismicnoiseisformedbycombiningthefourselectionmethods.Noiserecordsof85seismicstationsinFujianProvinceinJuly2013areselectedforverification,andtheresultsshowthattheanomalousnoise-recognitionsystemcouldreacha90%successrateatmoststationsandtheeffectofselectionareverygood.Therefore,itcouldbeappliedtotheseismicnoisereal-timemonitoringinstations.
简介:Terahertz(THz)radiation,whosefrequencyrangesfrom0.1THzto10.0THz,hasrichscience,butlimitedtechnology.Ithaslongbeenconsideredthelastremainingscientificgapintheelectromagneticspectrum.Farfrombeingfullyexploited,itoffersgreatopportunitiesinscience,innovation,newtechnology,andpotentialapplications.
简介:Inthispaper,weconsiderthestationaryprobabilityandfirst-passagetimeofbiasedrandomwalkon1Dchain,whereateachstepthewalkermovestotheleftandrightwithprobabilitiespandqrespectively(0p,q1,p+q=1).Wederiveexactanalyticalresultsforthestationaryprobabilityandfirst-passagetimeasafunctionofpandqforthefirsttime.Ourresultssuggestthatthefirst-passagetimeshowsadoublepower-lawF~(N-1)~γ,wheretheexponentγ=2forN<|p-q|~(-1)andγ=1forN>|p-q|~(-1).Ourstudyshedsusefulinsightsintothebiasedrandom-walkprocess.
简介:ThetraditionalorientedFASTandrotatedBRIEF(ORB)algorithmhasproblemsofinstabilityandrepetitionofkeypointsanditdoesnotpossessscaleinvariance.Inordertodealwiththesedrawbacks,amodifiedORB(MORB)algorithmisproposed.Inordertoimprovetheprecisionofmatchingandtracking,thispaperputsforwardanMOKalgorithmthatfusesMORBandKanade-Lucas-Tomasi(KLT).ByusingKalman,theobject’sstateinthenextframeispredictedinordertoreducethesizeofsearchwindowandimprovethereal-timeperformanceofobjecttracking.TheexperimentalresultsshowthattheMOKalgorithmcanaccuratelytrackobjectswithdeformationorwithbackgroundclutters,exhibitinghigherrobustnessandaccuracyondiversedatasets.Also,theMOKalgorithmhasagoodreal-timeperformancewiththeaverageframeratereaching90.8fps.