基于邊緣結(jié)構(gòu)幾何劃分的圖像檢索方法.pdf

1、中國海洋大學(xué)碩士學(xué)位論文基于邊緣結(jié)構(gòu)幾何劃分的圖像檢索方法姓名：賈小盟申請學(xué)位級別：碩士專業(yè)：信號與信息處理指導(dǎo)教師：王國宇2003.6.1苧主望絲竺塑些塑型坌塑塑堡竺壅—————————————一ImageRetrievalBasedonGeometricPartitioningforEdgeStructureAbstractThispaperprovidesanewimageretrievalapproach：Imageretrie

2、valbasedongeometricpartitioningforedgestructureThisapproachrepresentsimagefeaturesby“EdgePixelSegment(EPS)”，andestablishestheindexusingmaturetext—basedinformationretrievaltechnologyafterblockcodingtoedgeimageThespiritabo

3、utthisimageretrievalapproachderivesfromtext—basedIRTherearetwomajorproblemstosettleinordertogeneralizetext—basedIRtoimageretrievalOneiswhatare“keywords”ofanimageandhowtogeneratethem，theotherishowtOestablishtheindexusingt

4、hetextbasedindexingmodelsintheimagedomain“Edge—pixelSegment(EPS)”proposedinthispaperisanimportantconcepttepresentedgestructurefeaturesItisgeneratedbymeansofgeometricpartitioningforedgeimageanddescribedbyeigenvectonThisne

5、wimagerepresentationmethodhasmanyadvantagesItcarlreducegray—levelredundancybroughtbypartitioninggray—levelimagedirectlyandCanavoidthedifficultyofimagesegmentationandobject—orienteddescriptionssemanticallyItisimportanttha

6、tithasinvarianceofrotationandtranslationClusteringtheeigenvectorsofedgeimagesintrainingsettogeneratethecodebook，‘‘KeyEPSs”，encodingtheedgeimagesinimagedatabasebyvectorquantization，theimagesindatabaseandtheretrievalimageC

7、anberepresentedbyonedimensiondata，whichissimilartothetextdatastructureAtthistime，text—basedIRtechnologyCallbeusedtoimageretrievalThroughBooleanModelofinverted‘fileandHistogramModel，anewindexisestablishedaimedatsyntactica