Face image super-resolution through locality-induced support regression | |
Junjun Jiang; Ruimin Hu; Chao Liang; Zhen Han; Chunjie Zhang | |
发表期刊 | Signal Processing |
2014 | |
期号 | 103页码:168-183 |
摘要 | In thispaperweproposeanovelfaceimagesuper-resolution(SR)methodnamed Locality-inducedSupportRegression(LiSR).Givenalow-resolution(LR)inputpatch,we learnamappingfunctionbetweenthelocalsupportLRandhigh-resolution(HR)patch pairs topredictitsHRversion.ThesupportcanbeobtainedfromtheLRorHRpatch manifolds,whichleadstotwovarietiesofLiSR,namelyLRpatchguidedLiSR(LR-LiSR)and HR patchguidedLiSR(HR-LiSR).LR-LiSRdirectlylearnsthemappingfunctionbetween local supportLR/HRpatchpairsgivenaninputLRpatch.AsforHR-LiSR,thesupportanda mappingfunctionwillbeiterativelylearnedtoupdatethetargetHRpatch.Thekey advantagesofourproposedframeworkaretwo-fold:(1)thestrongregularizationof “same representation” of priorwork [1,2] is relaxedtothesamesupport,andhencemuch flexibilitycanbegiventothelearnedmappingfunction;(2)wedefinethesupportinthe LR orHRpatchmanifoldspacebyincorporatingthelocalityconstraint,whichcanwell preserve themanifoldstructureofthetrainingset.Experimentalresultsreportedonboth simulatedLRfaceimagesandreal-worlddatasetsdemonstratetheeffectivenessofthe proposed method. |
关键词 | Super-resolution Face Image Support Regression Manifold Learning |
文献类型 | 期刊论文 |
条目标识符 | http://ir.ia.ac.cn/handle/173211/15393 |
专题 | 类脑智能研究中心 |
推荐引用方式 GB/T 7714 | Junjun Jiang,Ruimin Hu,Chao Liang,et al. Face image super-resolution through locality-induced support regression[J]. Signal Processing,2014(103):168-183. |
APA | Junjun Jiang,Ruimin Hu,Chao Liang,Zhen Han,&Chunjie Zhang.(2014).Face image super-resolution through locality-induced support regression.Signal Processing(103),168-183. |
MLA | Junjun Jiang,et al."Face image super-resolution through locality-induced support regression".Signal Processing .103(2014):168-183. |
条目包含的文件 | 下载所有文件 | |||||
文件名称/大小 | 文献类型 | 版本类型 | 开放类型 | 使用许可 | ||
1-s2.0-S016516841400(5003KB) | 期刊论文 | 作者接受稿 | 开放获取 | CC BY-NC-SA | 浏览 下载 |
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。
修改评论