《刘震昌个人经历》PPT课件.ppt

劉震昌個人經歷 學歷博士 台灣大學電機工程所1998 碩士 交通大學資訊工程所1994 1996學士 交通大學資訊工程系1990 1994工作經歷中研院資訊所研究助理 國防役 1996 10 指導教授中研院資訊所黃文良副研究員台大電機系陳銘憲教授 論文著作 期刊論文1996IEEEtrans onConsumerElectronic AnewintensitystereocodingschemeforMPEG1audioencoder LayerIandII 指導教授 交大資工劉啟民教授 2000IEEEtrans onImageProcessing Estimationof2 DnoisyfractionalBrownianmotionanditsapplicationsusingwavelets SubmittedtoIEEEtrans onImageProcessing Ablock basedchannel optimizedsourcecodingforEZW likeprogressivecodecusingrate distortionanalysis 與中原大學電機系黃文吉教授合作 SubmittedtoIEEEtrans onCircuitsandSystemsforVideoTechnology Activecontourmodelusingwaveletmodulusforobjectsegmentationandtrackinginvideosequences 與陽明大學微免所林奇宏教授合作 論文著作 2 會議論文1998IEEESymposiumonTime FrequencyandTime ScaleAnalysis Parameterestimationanddenoisingof2 DnoisyfractionalBrownianmotionusingnon orthogonalwavelet 1999IEEEInt lConf onImageProcessing Robustblock basedEZWimagecompressionwithchannelnoiseoptimizedrate distortionfunctions 2000IEEEInt lConf onImageProcessing Amulti channelchannel optimizedschemeforEZWusingrate distortionfunctions 2001IEEEInt lConf onImageProcessing Wavelet basedactivecontourmodelforobjecttracking 研究方向 Wavelet basedsignalprocessingEstimationoffractionalBrownianmotionRobustcompressedimagetransmissionoverIP wirelessnetworkObjecttrackingusingactivecontourmodel Estimationof2 DFractionalBrownianMotion fBm usingWavelets 2 DfBm examples a 0 2 a 0 5 a 0 8 2 DIsotropicFractionalBrownianMotion Non stationarystochasticmodelfornaturalscenesandtextures1 fspectruma scalingexponent 0 a 1 whichisrelatedtofractaldimensionandsurfaceroughnessStatisticalself similarpropertyNetworktrafficmodelisshowntoexhibitself similarproperty ParameterEstimation Wavelettransformof2 DfBmiswide sensestationaryateachscale Varianceofwavelettransformof2 Dfbmfollows Application de nosing original original randomnoise de noised Application coastlinedetection threshold de noising Application texturesegmentation Application texturesegmentation 2 fBmmosaic AMulti ChannelChannel OptimizedSchemeforEZWusingRate DistortionFunctions Jen ChangLiu NationalTaiwanUniversityWen LiangHwang IIS AcademiaSinica TaiwanWen JyiHwang Chung YuanChristianUniversityMing SyanChen NationalTaiwanUniversity ReviewofEZWcoder EZW EmbeddedZerotreeWavelet State of the artlowbitratestillimagecoderproposedbyShapiroin1993 Embedded encodingsofthesameimageatlowerbitrateareembeddedinthebeginningprotionofthebitstreamforthetargetbitrate Wavelet discretewavelettransformZerotree groupingofinter bandcoefficientsbyzerotreesymbol NoisyChannel ProblemofEZWTransmissionthroughNoisyChannel Communicationmodel Sourcecoding Sourcedecoding source output EZW ProblemofEZWthroughNoisyChannel cont DuetohighcorrelationintheEZWencodedbitstream singlebiterrorwillpropagatetothewholeimage EZWBitstream error Hereafterarefalsedecoded Example DecodingofCorruptedEZWbitstream Originalimage Decodedimagethroughanoisychannelwithbiterrorrate 10 3 Solution Channel OptimizedSourceCoding channel optimizedsourcecoding Sourcecoding Sourcedecoding NoisyChannel source output EZW Channelproperty Proposed Block BasedMulti ChannelFramework SolutionStep1 Block basedEZW Idea block basedimagecodingwillincreaserobustness localizetheerroreffect Originalimage Divisionintoblocks SolutionStep2 BitAllocationwithRate DistortionConstraints Problem howtoallocatebitstoeachblock Idea usingrate distortionfunctionofeachblock BitAllocationwithRate DistortionConstraints cont Optimalbitallocationusingrate distortionconstraints Dynamicprogrammingcanbeusedtofindtheoptimalsolution whereDi bi istherate distortionfunctionforblocki andRisthetotalbitbudget SolutionStep3 AddNoisePropertyintoRate DistortionFunction Problem howdoweconsiderthenoisychannelproperty Solution takingnoiseparametersintherate distortionfunction Case memorylessBinarySymmetricChannel BSC 0 1 0 1 p p 1 p 1 p biterrorrate BER p AddNoisePropertyintoRate DistortionFunction cont Statisticalrate distortionfunctionforsomeblock Probabilitythatthefirsterroroccursatthei thbit Distortionwith i 1 correctbits Givennbits theexpectedvalueofdistortion AddNoisePropertyintoRate DistortionFunction cont Example ExperimentResult cont CorruptedEZW Ourapproach BSCnoisewithBER 10 3 ExperimentResult cont CorruptedEZW Ourapproach Burstnoise Wavelet basedActiveContourModel Snake forObjectTracking Jen ChangLiu Wen LiangHwangInstituteofInformationScience AcademiaSinica TaiwanMing SyanChenNationalTaiwanUniversity TaiwanJin WuTsai Chi HungLinInstituteofMicrobiologyandImmunology NationalYang MingUniversity Taiwan Problem Givenanimagesequence video trackthemotionofcertainobjectApplication biologicalcelltracking ActiveContourModel Snake Processflow Manualinitialization inputfirstframe Contourfinding Motionestimationfornextframe a b a b c c d d snake DesignIssues Activecontourmodel Snake Multi resolutionapproachDesignofpotentialfieldMotionestimationWaveletdomainapproachBothsolutionsusethewaveletmodulus Snake Theory Parametriccontourfunction v s x s y s s 0 1 OptimizationproblemFindv s thatminimizeE Howtodesignthis Potentialfield Example Potentialfield 3Dplot Multi resolutionapproach Wavelet basedSnake Takethemodulusmaximumofthehigh passwaveletcoefficientsaspotentialfield Multi resolutionapproach Wavelet basedSnake 2 WhyModulusMaximum gradientofsmoothedimageisusuallytakenaspotentialfield Wedefinepotentialfieldas wavelettransforminx andy direction Multi resolutionapproach Wavelet basedSnake 3 Scale1 potentialfield Scale2 potentialfield Scale3 potentialfield Scale4 potentialfield Originalimage Wavelettransform MotionEstimation Resultingobjectcontourofframet Initialobjectcontourofframet 1 Motionestimation Theneedformotionestimation provideainitialcontourfornextframe MotionEstimation 2 Waveletdomainapproach MotionEstimation 3 previouscontour motionestimation scale22smoothedimage scale22waveletmodulus scale21waveletmodulus refinement scale20finalresult CompleteFrameworkusingWaveletModulus BothSnakeandmotionestimationusewaveletmodulusforcalculation ExperimentResult Frame1 Frame12 Frame25 result 目前研究題目 Robustcompressedimagetransmission ARQ basedDiversitySystem QoSguarantee ARQMultiplechanneltransmission Diversity