基于分数阶运动估计的小波视频编码(IJIGSP-V7-N8-8)

I.J.Image,GraphicsandSignalProcessing,2015,8,66-75PublishedOnlineJuly2015inMECS()DOI:10.5815/ijigsp.2015.08.08Copyright©2015MECSI.J.Image,GraphicsandSignalProcessing,2015,8,66-75Wavelet-basedVideoCodingusingAdvancedFractionalMotionEstimationTechniqueWissalHassenandHamidAmiriSignal,ImageandInformationTechnologieslaboratoryTheElectricalEngineering[e-mail],[e-mail]—Thepurposeofthispaperistoencodeacolorvideobywavelettransformation.Therefore,weproposeanewhybridapproachwhichcombinesafractionalmotionestimationtechnique.Severalstudieswerecarriedouttoreducethespatialandtemporalredundancies,henceatthelevelofspatialvideocoding,weuseanewapproachbasedonsub-bandscodingthroughadiscretewavelettransformation.ThistechniqueisbasedontheprincipleoftheEZWalgorithmofShapiro.Itproceedsbyseparatingtheencodingofthesignsandthemagnitudesofwaveletcoefficients.Then,attheleveloftemporalcompression,weproposeastudyofmotionestimationwithdifferentaccuracybasedonimageinterpolationtoimprovethequalityofpredictedframe.Next,wepresentarepresentationreducingthesizeofthemotionvectorfieldandwecompressitbytwoofentropiccodingapproachesnamelyHuffmancodingandarithmeticcoding.Theproposedvideocodecwasappliedonavideosequencewithdifferentsizes(CIFandQCIF)anddifferentdynamics.Theobtainedresults,intermsofobjectiveassessment(PSNR,theSSIMandVQM),weresatisfactorycomparedwithothervideocodingstandards.WehavealsoproposedasubjectiveevaluationandtheresultsarecomparedtothoseobtainedbyH.264/AVCstandard.IndexTerms—Wavelettransform,H.264/AVCstandard,imagequalityassessment,fractionalmotionestimation,videocoding.I.INTRODUCTIONVideocompressionisnecessaryforstorageandtransmissioninmultimediaapplications.Inavideosequencetherearetwokindsofredundancy;thespatialredundancyrepresentedbyblocksofpixelsthatrepeatinthesameimageandtemporalredundancythatoccurswhenblocksofpixelsarepermanentintwoormoresuccessiveimagessuchasthecaseofafixedbackground.Theobjectofthevideocompressionistoreducethoseredundancies.Thefirstpartofthispaperisreservedforintra-framecoding.Then,inthesecondpart,wepresentthechoseninter-framecompressiontechniqueandfinally,wepresenttheoverallvideocodingalgorithmanditsresultswithobjectiveandsubjectiveevaluations.Thespatialcompressionisbasedontheimageencodinginafrequencydomain.Therefore,theimagegoesthroughthreemainstages:frequencytransformation,quantificationandanencodingstep.ThestandardofimagecompressionJPEGusestheDiscreteCosineTransform(DCT).Thistransformationisappliedonblocksofpixels;thereforethedecodedimagepresentsartifactsintheseblocks.However,JPEG.2000standardusesDiscreteWaveletTransform(DWT)whichisaverypowerfultoolforimagecompressionandprovidesbothspatialandfrequencylocalizationofimageenergy.DuetothegoodqualityoftheDWT-basedimagecoding,severalstudieshaveusedthistechniquesuchastheEmbeddedZerotreeWaveletalgorithm(EZW)ofShapiro[1]andtheSetPartitioningInHierarchicalTrees(SPIHT)codingalgorithm,proposedbySaidandPearlman[2].Inourwork,wepresentanewwaveletbasedcodecinspiredfromtheEZWalgorithmandbasedonseparateentropycodingofsignandmagnitudeofwaveletcoefficients.ThenweshowthatthisalgorithmprovidesaverygoodqualityofcompressionforvideoframesbyevaluatingitsperformancecomparedtoJPEG.2000standard.Thechoicesofwaveletdecompositionlevelaswellastheproperwaveletfilterarediscussedinthissection.Themainobjectofthesecondpartofthispaperistoreducethetemporalredundancies,thenwepresenttheprincipleofmotionestimationbyBlockMatchingandweimprovetheperformanceofthistechniquebyintroducinganewfractionalestimationtechniquebasedonimageinterpolation.Thereafter,weproposeacodingtechniqueofmotionvectorbyentropycodingandcomparetheresultsfoundfortwocodingalgorithms:Huffmanencoderandarithmeticencoder.Inthethirdpartofthispaper,weproposeanewvideocompressionschemewhichisbasedonourpreviousresults.Theproposedschemeprovidesverysatisfyingresultswhichareevaluatedrelativetoothervideocodingstandards.II.WAVELET-BASEDSPATIALVIDEOCODINGThespatialcodingalgorithmsarebasedonthedecompositionoftheimageintosub-bandsasshowninFig.1.Infactthesub-bandcodingbyDCTorDWTisinitiallyusedinspeechprocessing[3]andithasbeensuccessfullyextendedtostillimagecompression[4,5].AWavelet-basedVideoCodingusingAdvancedFractionalMotionEstimationTechnique67Copyright©2015MECSI.J.Image,GraphicsandSignalProcessing,2015,8,66-75sub-bandsencoderdecomposestheimageintosub-images.InDCT,theimageissplitinblocksofpixelsandthetransformationisappliedtoeachblockasanindependentsub-image.InthecaseofDWT,theimageisdecomposedinwaveletdomainwithoutblocksplitting.TheDCTisappliedbyblocksofpixelswhichproducethephenomenonofartifactsthatisnotthecasefortheDWTwhichisappliedtotheentireimage.Thereafter,eachsub-bandgoesthroughadifferentencodingdependingondesiredqualityandcompressionbitrate.Then,aprogressiveencodingbit-planebybit-planeisapplied.Fig.2showsanexampleofprogressivecodingandtheresultsforeachcompressionbitrate.ThevalueBpp(Bitsperpixel)indicatesthebitsrequiredtoencodeea