小区域破损图像修复算法与其应用-研究

《小区域破损图像修复算法与其应用-研究》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、AbstractDuetothedevelopmentofdigitalimagetechnologyinrecentyears，dⅢgtalimageinpaintingtechnologyhasbecomeanimportantresearchfieldof1mageprocessing。Accordingtothefieldofimageinpaintingapplication，itcanbedividedintotwoparts：imageinpaintingalgorithmforsmalldamagedscaleandfillingalgorithmforlarg

2、earea·Thisthesisselectsimageinpaintingalgorithmbasedonsmalldamagedscaletostudv，includingimageinpaintingalgorithmbasedonNo‘mallzedConvolution(NC)andtotalvariation(TV)．Onthebasisofstudyingtheexl8tlngnormalizedconvolutionimageinpaintingalgorithm，thesawtootheffectandthesuperpositionerrorhavebeen

3、found．ThereasonofsawtootheffectlSonlYuslngspatialinformarionwhencalculatingtheneighborhoodweightmatrix，butnotconsideringthefeatureinformationofimage．Inthisthesis，theneighborhoodweightmatrixofadamagedpixeltobeestimatedisconstructedbyCOraputmgtheisophoteinformation，toensuretorepairthedamagedar

4、eaalongthelsoPhotesdirection．Thecauseofsuperpositionerroristhattheimagerestorationalgorithmbasedoninterpolationisagreedyalgorithm．InthealgorithmthepixelwhichbeinpaintedDriorhaslargeinfluenceonthepixelswhichbeinpaintedaftel·lhewindowradiusofdamagedpixeltobeestimatedisadaptivelyselectedacco‘dl

5、ngtothedistancebetweenthedamagedpixelandthedamagedboundary，andtheproportionofimageknowninformationisincreased，toreducethesuperposltlonerror．Experimentalresultsshowthattheboundaryandtextureinformation．proposedalgorithmcallbettermaintainIma叠einpaintingalgorithmbasedOntotalvariation(TV)isadiffu

6、sion-basedaIgorithm．Thebasicideaistodiffusiontheneighborhoodinformationotdamagedareatothedamagedarea．Thisalgorithmissensitivetotheselectionofparameters，andtheconvergencespeedisslow．ThisthesisproposesanadaptlVe1mageinpaintingalgorithmbasedontotalvariation．Thevalueofparamate¨88electedadaptivel

7、ybytheneighborhoodinformationofpixelwhichtobeestimated·Inthequalityofinpaintingcondition，thisalgorithmhasthefasterconVergencesPeedthantotalvariation．Atthesametime，inthestudyofTVmodelalgorithm，smootheflfectwasfound．ThesmoothingeffectofTVmodelalgorit