面向網(wǎng)絡(luò)圖像過濾的文本抽取技術(shù)研究.pdf

1、南開大學(xué)碩士學(xué)位論文面向網(wǎng)絡(luò)圖像過濾的文本抽取技術(shù)研究姓名：李嬌鳳申請學(xué)位級別：碩士專業(yè)：計(jì)算機(jī)應(yīng)用技術(shù)指導(dǎo)教師：王慶人201205AbstractAbstractWiththerapiddevelopmentofInternetandgeneralizationofhandhelddigitaldevices，therearemoreandmoredigitalimagesinthenetworkAvarietyofapplicati

2、onsappearatthesametimeTheseapplicationsincludeimageretrieval，imagestorageandmanagementandSOonTheimagesonIntemetbringmuchusefulinformationforUS，butatthesametimesomenegative，badinformationalsoexistsintheseimagesThecommonfo

3、rmsoftheseimagesarepomographicimages，spamandbadimagesTheseobjectionableimageshaveharmfuleffectsonsocialstabilityandteenagers’healthygrowthSoithasgreatsignificancetomanagetheimagesonInternetandfilterthebadonesThetechnolog

4、iesofnetworkimagefilteringCanbedividedintothreecategories，includingURLblocking，imagefilteringbasedonkeywordsandimagefilteringbasedoncontentofimageContentbasedimagefilteringmethodismoreadaptiveandeffectivethantheothertwoo

5、nes，anditalsohaswiderapplicationsthanotherimagefilteringmethodsInthispaperweusecontentbasedimagefilteringasourmethodInthispaperweproposeamulti—resolutiontextextractionmethodFirsttheimagepyramidisbuiltupwithfixedscale12We

6、usefixedwindowwithsize16木16slidingoneachlayerofthepyramidimageandextractHOGfeaturesofeachwindow；thenwefeedtheHOGfeaturesintothetrainedAdaboostclassifiertojudgewhethercurrentwindowbelongstotextarea，andmergethetextareasofe

7、achdecomposedimagewithsimplerules；afterwegetthemergedtextarea，weuseCannyedgedetectorandSobelgradientoperatortoestimatetheaveragestrokewidthofthetextarea；atlastweproposeannovelimagebinarizationmethodbasedonwaveletfilterin

8、gandusethismethodtoseparatetextfrombackgroundoftextimageInthismethodthescaleofwaveletfilteringisselectedaccordingtothestrokewidthoftextWetestourmethodonpublicdatasetprovidedbyICDAR2003andrandomlyselect56imagesfromthedata