資源描述:
《特征選擇算法及其在基于內(nèi)容圖像檢索中的應(yīng)用研究》由會員上傳分享,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。
1、重慶大學(xué)博士學(xué)位論文特征選擇算法及其在基于內(nèi)容圖像檢索中的應(yīng)用研究姓名:李云申請學(xué)位級別:博士專業(yè):計(jì)算機(jī)軟件與理論指導(dǎo)教師:吳中福;劉嘉敏20050301重慶大學(xué)博士學(xué)位論文算法,該類算法是建立在統(tǒng)計(jì)學(xué)習(xí)理論的基礎(chǔ)上,尋求結(jié)構(gòu)風(fēng)險(xiǎn)最小化的特征子集,主要是利用對支持向量機(jī)的性能影響進(jìn)行特征選擇,選擇的效果很好。目前的研究還只是對樣本類別已知的情況下進(jìn)行特征選擇,隨著支持向量機(jī)的理論研究不斷深入,支持向量機(jī)用于非監(jiān)督特征選擇是完全可能的。另外,還對特征選擇算法的應(yīng)用進(jìn)行了初步探討,并以特征選擇算法在基于內(nèi)容圖像檢索中的應(yīng)用作為例子,詳細(xì)分析了
2、基于內(nèi)容的圖像檢索中特征選擇的必要性和采用的方法,對其特有的方法——相關(guān)反饋技術(shù)進(jìn)行了深入的分析研究,給出其理論模型。同時(shí)將前面提出的監(jiān)督高維特征選擇算法在圖像數(shù)據(jù)庫中做了粗略的實(shí)驗(yàn),獲得不錯(cuò)的效果。此外特征選擇還廣泛應(yīng)用于文本分類、入侵檢測、基因分析等,隨著機(jī)器學(xué)習(xí)、數(shù)據(jù)挖掘和模式識別領(lǐng)域的不斷擴(kuò)大,特征選擇算法的應(yīng)用領(lǐng)域也將擴(kuò)展。本文最后對研究工作進(jìn)行了總結(jié),提出了今后進(jìn)一步的研究方向。關(guān)鍵詞:特征選擇算法,模糊集,支持向量機(jī),基于內(nèi)容的圖像檢索II英文摘要ABSTRACTWiththescopeandfieldsofcomputera
3、pplicationexpandedincreasingly,andinparticulartherapiddevelopmentoftheInternet,largeevenhugeamountofdatahasbeenproducedinvariousapplicationsystemsandontheInternet,resultingintheproblemandphenomenonof“dataexplosionandknowledgescarcity”;dataminingisthemosteffectivemethodtota
4、ckletheproblem.However,datapreprocessingisessentialfactorforeffectivedatamining,andfeatureselectionisoneofthemostimportantdatapreprocessingmethods.Inaddition,featureselectionisnecessarystepformachinelearningandpatternrecognition.Theresearchoffeatureselectionstartedfrom60’s
5、lastcenturywithmanyachievements.However,withtheappearanceofnewapplicationdomainsandobjects,therearestillmanyproblemsshouldbesolvedforfeatureselectionurgently.Thispapergivesadetailedintroductiontothese,andmakesin-depthresearchoncurrentfocus,especiallythealgorithmsoffeatures
6、election,withcertainproducts.Theauthordividestheresearchoffeatureselectionalgorithmintothreestages,atfirst,putsforwardthemodelofcommonfeatureselectionalgorithm.Atthesametime,fromtheresearcheranduser’sperspective,categorizesthefeatureselectionalgorithms.Thesewillfacilitatet
7、heusertoselectappropriatealgorithm,promotetheapplicationandbuildsolidbasefortheresearchofit.Secondly,presentsandintroducessomespecificalgorithmsoffeatureselection,whicharethefocusandhotspotofcurrentresearch.Theycontain:algorithmsoffeatureselectionfromfuzzyfeaturespace,from
8、highdimensionalfeaturespace(supervisedandunsupervised)andalgorithmsoffeatureselectionusin