简介:Watershedtransformationisapowerfulmorphologicaltoolimagesegmentation.However,theperformanceoftheimagesegmentationmethodsbasedonwatershedtransformationdependslargelyonthealgorithmforcomputingthegradientoftheimagetobesegmented.Inthispaper,wepresentamulti-scalegradientalgorithmbasedonmorphologicalopenatorsforwatershed-basedimagesegmentation,witheffectivehandlingofbothstepandblurrededges.Wealsopresentanalgorithmtoeliminatethelocalminimaproducedbynoiseandquantizationerrors.Experimentalresultsindicatethatwatershedtransformationwiththealgorithmsproposedinthispaperproducesmeaningfulsegmentations,evenwithoutaregion-mergingstep.
简介:Fromtheperspectiveofcompressedsensing(CS)theory,thechannelestimationprobleminlarge-scalemultipleinputmultipleoutput(MIMO)-orthogonalfrequencydivisionmultiplexing(OFDM)systemisinvestigated.Accordingtothetheory,thesmallermutualcoherencethereconstructionmatrixhas,thehighersuccessprobabilitytheestimationcanobtain.Aimingtodesignapilotthatcanmakethesystemreconstructionmatrixhavingthesmallestmutualcoherence,thispaperproposesalowcomplexityjointalgorithmandobtainsakindofnon-orthogonalpilotpattern.Simulationresultsshowthatcomparedwiththeconventionalorthogonalpilotpattern,applyingtheproposedpatternintheCSchannelestimationcanobtainthebetternormalizedmeansquareerrorperformance.Moreover,thebiterrorrateperformanceofthelarge-scaleMIMO-OFDMsystemisalsoimproved.