深度图增强算法分析及其在自由视点视频中的应用-analysis of depth map enhancement algorithm and its application in free view video.docxVIP

  • 8
  • 0
  • 约4.09万字
  • 约 53页
  • 2018-07-31 发布于上海
  • 举报

深度图增强算法分析及其在自由视点视频中的应用-analysis of depth map enhancement algorithm and its application in free view video.docx

深度图增强算法分析及其在自由视点视频中的应用-analysis of depth map enhancement algorithm and its application in free view video

AbstractThevigorousdevelopmentsofcomputervisiontechnology,videotechnologyandnetworktechnologyprovidethepossibilityforthedevelopmentoffreeviewpointvideo(FVV).TheoneofthekeypointsofFVVishowtorepresent3Dgeometryofrealscene(3Drepresentation).Atpresent,thereexisttwokindsof3Drepresentations:a)Multi-viewbased3Drepresentation;b)depthimagebased3Drepresentation.Theformercanrenderhigh-qualityimageofvirtualviewpointbycapturing3Dinformationwithcameraarray.However,thewayofhigh-densitysamplingconsumesalotresourceofcodecsandnetworkbandwidth.Thesecondsolutionusescolortextureplusdepthimagerepresenting3Dinformationofrealscene,whichgreatlyreducesthebandwidthrequirementsandthereisnoneedofhigh-densitysamplingduetodepthimagecandescribethe3Dgeometry.Additionally,themethodcangeneraterealisticimageofarbitraryvirtualviewpointinalimitedrange.Therefore,Depthimagebased3DrepresentationisgenerallyconsideredasthealternativeschemeaboutthenextgenerationofFVVduetoadvantagesoflownetworkbandwidth,highspeedofvirtualperspectivedrawingandrealisticrendering.However,obtaininghigh-qualitydepthisquitechallenging.Toaddressthisproblem,ourpapersfocusondepthimageenhancementalgorithmresearchandachievedthefollowingresults:Firstly,weproposedadepthimageenhancementalgorithmbasedonimprovingbilateralfiltering.Thecurrentdepthofimagehastwomainmethods:passiveopticalmethodsandactivesensor.Thepassivemethodistoobtaindepthimagethroughcalculatingthefeaturescorrespondingrelationshipbetween2Dimageand3dgeometrybycomputervisiontechnology.Passiveopticalmethodrequireshightimecomplexity,strictconstraintsandaccurateimagecorrectionetc.Thereforeitisdifficulttobeappliedinpractice.Activeopticalmethodcapturesthedepthinformfrom3Drealsceneusingtime-of-flight(TOF)orlightcoding.Comparedtopassivemethod,activemethodbecomesanincreasinglyconcernedmethodduetoitsreal-time,accuracyandpracticality.However,thelow-costsensorhasthedrawbackoflowresolution,cavity,unmatchedboundaryofedge,limitingitsapplicationinpractice.Therefore,thepaperusesjointbilateralfil

您可能关注的文档

文档评论(0)

1亿VIP精品文档

相关文档