資源描述:
《基于互補(bǔ)特征的紋理圖像檢索》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在行業(yè)資料-天天文庫。
1、基于互補(bǔ)特征的紋理圖像檢索摘要:針對(duì)互補(bǔ)特征可以有效地改善圖像檢索系統(tǒng)性能的特點(diǎn),提出一種在改進(jìn)contourlet變換域采用l1能量與廣義高斯分布參數(shù)特征的紋理圖像檢索方法。首先,應(yīng)用改進(jìn)的方法對(duì)方向子帶系數(shù)進(jìn)行廣義高斯統(tǒng)計(jì)建模。然后,分別單獨(dú)利用各個(gè)特征和相應(yīng)的相似性測度進(jìn)行檢索。最后,基于直接的相似性測度和,采用這兩種互補(bǔ)的特征進(jìn)行檢索。實(shí)驗(yàn)結(jié)果表明,和采用單一特征相比較,互補(bǔ)特征由于充分地反映了圖像的結(jié)構(gòu)信息和隨機(jī)分布信息,從而有效地提高了紋理圖像數(shù)據(jù)庫的平均檢索率。關(guān)鍵詞:改進(jìn)的contourlet變換;建模;l1能量
2、;廣義高斯分布;互補(bǔ)特征;紋理圖像檢索textureimageretrievalbasedoncomplementaryfeaturesquhuai.jing*schoolofinformationandelectricengineering,shandongjianzhuuniversity,jinanshandong250101,chinaabstract:becausetheperformanceoftheimageretrievalsystemcouldbeeffectivelyimprovedusingthecomplem
3、entaryfeatures,aretrievalmethodofthetextureimageusingl1energyandgeneralizedgaussiandistributionparameterfeatureswasproposedintheimprovedcontourlettransformdomain.firstly,thedirectionalsubbandcoefficientsweremodeledasgeneralizedgaussiandistributionusinganimprovedapproach
4、.then,thetextureimageswererespectivelyretrievedbasedonthesinglefeatureandthecorrespondingsimilaritymeasurement.lastly,usingthecomplementaryfeaturesandthedirectsummationoftheirsimilaritymeasurements,thetextureimageswereretrieved.theexperimentalresultsshowthat,comparedwithsi
5、nglefeature,theaverageretrievalratesofthetextureimagedatabaseareeffectivelyimprovedbythecomplementaryfeaturesduetofullyrepresentingthestructuralinformationandtherandomdistributioninformation.becausetheperformanceoftheimageretrievalsystemcouldbeeffectivelyimprovedbyusingthe
6、complementaryfeatures,aretrievalmethodofthetextureimageusingl1energyandgeneralizedgaussiandistributionparameterfeatureswasproposedintheimprovedcontourlettransformdomain.firstly,thedirectionalsubbandcoefficientswentthroughgeneralizedgaussianmodelingwithanimprovedapproach
7、.then,thetextureimageswererespectivelyretrievedbasedonthesinglefeatureandthecorrespondingsimilaritymeasurement.lastly,usingthecomplementaryfeaturesandthedirectsummationoftheirsimilaritymeasurements,thetextureimageswereretrieved.theexperimentalresultsshowthat,comparedwithsi
8、nglefeature,theaverageretrievalratesofthetextureimagedatabaseareeffectivelyimprovedbythec