資源描述:
《基于PDE、圖割的圖像分割模型及算法碩士論》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫(kù)。
1、河南大學(xué)碩士學(xué)位論文基于PDE、圖割的圖像分割模型及算法姓名:劉玉霞申請(qǐng)學(xué)位級(jí)別:碩士專業(yè):應(yīng)用數(shù)學(xué)指導(dǎo)教師:宋錦萍;臺(tái)雪成20090501河南人學(xué)碩士學(xué)位論文中文摘要圖像分割是進(jìn)一步進(jìn)行圖像識(shí)別、分析和理解的基礎(chǔ),已有許多分割方法但還沒有一種普遍適用于各種圖像的有效方法.圖像分割的研究是目前圖像處理中研究的熱點(diǎn)之一.因此,本文基于偏微分方程(PDE)、圖割的圖像分割模型、算法研究具有理論意義和應(yīng)用價(jià)值.首先,本文從圖像分割的目的和意義出發(fā),對(duì)Mumford—Shah(M-S)模型、CV模型、逐段常水平集圖像分割
2、模型簡(jiǎn)單介紹和分析.針對(duì)MSWD模型數(shù)值計(jì)算算法效率低、穩(wěn)定性條件難以控制、調(diào)節(jié)參數(shù)難的問題進(jìn)行研究.提出算子分裂圖像分割模型,研究模型的數(shù)值計(jì)算,利用算子分裂格式給出了MSDM的AOS(AdditiveOperatorSplitting),AOS—Newton(AdditiveOperatorSplitting-Newton)兩種快速算法,從而達(dá)到既快速極小化模型又克服了時(shí)間步長(zhǎng)難以選擇的難題,參數(shù)調(diào)節(jié)范圍增大.其次,給出了求解MSDM的圖割算法.將求能量函數(shù)的最小值轉(zhuǎn)化為求圖的最小割.采用高效的最大流算法進(jìn)行
3、數(shù)值計(jì)算.圖割算法避免了其它優(yōu)化算法局部最小的缺點(diǎn),參數(shù)調(diào)節(jié)簡(jiǎn)單,計(jì)算速度快.對(duì)合成圖像及真實(shí)圖像利用提出的算法實(shí)驗(yàn),驗(yàn)證該算法的可行有效性.最后,論文對(duì)提出的三種算法通過數(shù)值實(shí)驗(yàn)進(jìn)行進(jìn)一步比較.關(guān)鍵詞:圖像分害-I];PDE;AOS;圖割河南人學(xué)碩士學(xué)位論文AbstractImagesegmentationistofurthercarryoutimagerecognition,analysisandunder-standing.thereareanumberofsegmentationmethods.Butth
4、ereisnotamethodwhichisgenerallyapplicabletoallkindsofimage.Therefore,thestudyofimagesegmentationisoneofthehotspotsintheresearchofimageprocessing.Theresearchofimagesegmen-tationbasedonpartialdifferentialequations(PDE)andgraphcutismeaningfulbothintheoryandpract
5、ice.Firstofall,inthispaper,summarizethepurposeandsignificanceofimagesegmen-tation,introduceandanalyzeMumford—Shahmodel,Chan-Vesemodel,piecewiseconstantlevelsetmethod.ThenumericalalgorithmforthemodelisinefficientanddifficulttoCOIl-trolthestabilityofconditions,
6、adjustingtheparametersoftheproblemdifficult.Tosolvethoseproblems,wegivetheoperatorsplittingimagesegmentationmodel.Usingofoper—atorsplittingscheme,wegivetwofastalgorithms--AOS(AdditiveOperatorSplitting),AOS—Newton(AdditiveOperatorSplitting—Newton),SOastoachiev
7、enoto出minimizeenergyfunctionquicklybutalsoovercomethedifficultiesinchoosingthetimesteps.Theparametercanbechosenmoreeasily.Secondly,usinggraphcutstechnologytosolvethegivenMSDM.Insteadofmini-mizingtheenergyfunction,weproposetosolvetheresultingminimizationproble
8、mbygraphcuts.Byfindingtheminimumcutonaspecialgraph,weobtainthesolutionforthesegmentationproblem。Highlyefficientmaximumflowalgorithmcanavoidtheshortcom—ingsoflocalminimum.Furthermore,param