基于小波、輪廓波和ICA的圖像基學(xué)習(xí).pdf

1、獨(dú)創(chuàng)性聲明本人所呈交的學(xué)位論文是在導(dǎo)師指導(dǎo)下進(jìn)行的研究工作及取得的成果。盡我所知，除特別加以標(biāo)注的地方外，論文中不包含其他人的研究成果。與我一同工作的同志對(duì)本文的研究工作和成果的任何貢獻(xiàn)均已在論文中作了明確的說(shuō)明并已致謝。本論文及其相關(guān)資料若有不實(shí)之處，由本人承擔(dān)一切相關(guān)責(zé)任。燃者簽名：：；多互旦堡壘≥。哆年3月2(7日學(xué)位論文使用授權(quán)本人作為學(xué)位論文作者了解并愿意遵守學(xué)校有關(guān)保留、使用學(xué)位論文的規(guī)定，即：在導(dǎo)師指導(dǎo)下創(chuàng)作完成的學(xué)位論文

2、的知識(shí)產(chǎn)權(quán)歸西安理工大學(xué)所有，本人今后在使用或發(fā)表該論文涉及的研究?jī)?nèi)容時(shí)，會(huì)注明西安理工大學(xué)。西安理工大學(xué)擁有學(xué)位論文的如下使用權(quán)，包括：學(xué)?？梢员４鎸W(xué)位論文；可以采用影印、縮印或其他復(fù)制手段保存論文；可以查閱或借閱。本人授權(quán)西安理工大學(xué)對(duì)學(xué)位論文全部?jī)?nèi)容編入公開(kāi)的數(shù)據(jù)庫(kù)進(jìn)行檢索。本學(xué)位論文全部或部分內(nèi)容的公布(包括刊登)授權(quán)西安理工大學(xué)研究生學(xué)院辦理。經(jīng)過(guò)學(xué)校保密辦公室確定密級(jí)的涉密學(xué)位論文，按照相關(guān)保密規(guī)定執(zhí)行；需要進(jìn)行技術(shù)保密的學(xué)

3、位論文，按照《西安理工大學(xué)學(xué)位論文技術(shù)保密申請(qǐng)表》內(nèi)容進(jìn)行保密(附《西安理工大學(xué)學(xué)位論文技術(shù)保密申請(qǐng)表》)。保密的學(xué)位論文在解密后，適用本授權(quán)。論文作者簽名：高一勺鋤導(dǎo)師杉簽名：私考肋侈年弓月功日AbstractTitle：IMAGEBASESLEARNlNGBASEDONTHEWAVELET，CONTOURLETANDlCAMajor：ComputationalMathematicsName：XiangyangGAOSuperviso

4、r：ProfFangDAIAbstractsignatu陀：塑幽趲s啪atu聯(lián)嗚逝Imagebaseslearning，whichasakindofimportantmethodofimagefeatureextractionandimagerepresentation，showsgreatsuperiorityTheimagewasdecomposedintosimpleimageprimitiveform，whichnotonlyc

5、anreducedtheinformationredundancy，butalsocancompressedtheimagedimensions，andbenefittheimagemodelingImagemodelingisanindispensablestepinimagerecognitionandsegmentationInthisthesis，westudiedthestatisticalpropertiesofcoeffi

6、cientofwaveletandcontourletswhichobtainedbywavelettransformandcontourletstransformforimage，andthentheIndependentComponentAnalysis(ICA)wasutilizedtolearnimagebasesThecontentsareasfollows(1)Animagebaseslearningmethodbasedo

7、nwaveletandICAwasproposedThismethodWasmakingfulluseofthehighdegreeofenergycorrelationofwavelettransformwithpropagationofthescaleandtheinherentcharacteristicsrelatedcoefficientsinthesamescaleTheparametersoftheHiddenMarkov

8、TreemodelWascombinedinthismethodAspatiallyadaptivethresholdwasusedtoclassifythesubbandwhichcontainedthefeaturecoefficients，andthenfeaturecoefficientsofcorrespondingfrequencysubbandwereselectedThewaveletinversetransformWa

9、sappliedtoobtainfeatureblocksofthegivenimageFinally，theICAwasusedtoleamimagebasesforthesefeatureblocks(2)AnimagebaseslearningmethodbasedoncontourletsandICAWasproposedFourfeatureblockselectedschemeweregiveninthismethodThe

10、firstschemegavefullconsiderationtothefeatureofcoefficientsandtherelationshipbetweenthecoefficientsEachcoefficientwithinthegivenscaleWassetupawindowThecoefficientsinthewindowweretodothewavelettransformonceagainWecalculate

11、dtheratioofthesumofgrayentropyofhighfrequencysubbandpartandthesumofgrayentropyoftheothersubbandThefeatumareaunderthecurrentscaleWasdeterminedbytheaboveratio，andthenaseriesoffeatureareawereobtainedbyusingtherelationshipbe