Adaptive Sparsity Matching Pursuit Algorithm for Sparse Reconstruction

《Adaptive Sparsity Matching Pursuit Algorithm for Sparse Reconstruction》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、IEEESIGNALPROCESSINGLETTERS,VOL.19,NO.8,AUGUST2012471AdaptiveSparsityMatchingPursuitAlgorithmforSparseReconstructionHonglinWuandShuWangAbstract—Thisletterpresentsanewgreedymethod,calledwheredenotesthevectornormandisAdaptiveSparsityMatchingPursuit(ASMP),

2、forsparsesolutions-orderrestrictedisometryconstant.When,theofunderdeterminedsystemswithatypical/randomprojectionRIPimpliesthattheprojectionoperatornearlymaintainsthematrix.Unlikeanteriorgreedyalgorithms,ASMPcanextractdistancebetweeneachpairof-sparsesign

3、als[3].informationonsparsityofthetargetsignaladaptivelywithawell-designedstagewiseapproach.Moreover,ittakesadvantageVariouseffortshavebeenmadeonthefastnumericalofbacktrackingtorefinethechosensupportsandthecurrentmethodsforsparsereconstructionwithreliable

4、accuracyandapproximationintheprocess.Withtheseimprovements,ASMProbustnesstonoise.Amongtheexistingrecoveryalgorithms,providesevenmoreattractiveresultsthanthestate-of-the-arttwomajorapproachesare-minimizationandgreedyalgo-greedyalgorithmCoSaMPwithoutprior

5、knowledgeofthespar-rithms.-minimizationmethods,suchasBasisPursuit(BP)sitylevel.Experimentsvalidatetheproposedalgorithmworkswellforbothnoiselesssignalsandnoisysignals,withtherecovery[4]anditsoptimizationalgorithmGPSR[5],solveaconvexqualityoftenoutperform

6、ingthatof-minimizationandotherminimizationprobleminsteadofthecombinatorialproblem.greedyalgorithms.-minimizationmethodsworkcorrectlyforallsparsesignalsIndexTerms—Adaptivegreedyalgorithm,blindsparserecon-andprovidetheoreticalperformanceguarantees.However

7、,theystruction,compressivesensing.aresensitivetonoiseandoftensufferfromheavycomputationalcomplexitywhensolvingmassiverecoveryproblems.Ontheotherhand,greedyalgorithmsiterativelyidentifythesupportsI.INTRODUCTIONofthetargetsignalandconstructanapproximation

8、onthesetofthechosensupports,untilahaltingconditionismet.Theycansolvelarge-scalereconstructionproblemsmoreefficiently.TheOMPRESSIVEsensing(CS)[1],emergedasanalterna-earliertypicalgreedyalgorithmsincludeorthogonalmatchingCtiveapproa