摘要
Asimpleandeffectivegreedyalgorithmforimageapproximationisproposed.Basedonthematchingpursuitapproach,itischaracterizedbyareducedcomputationalcomplexitybenefitingfromtwomajormodifications.First,ititerativelyfindsanapproximationbyselectingMatomsinsteadofoneatatime.Second,theinnerproductcomputationsareconfinedwithinonlyafractionofdictionaryatomsateachiteration.Themodificationsareimplementedveryefficientlyduetothespatialincoherenceofthedictionary.Experimentalresultsshowthatcomparedwithfullsearchmatchingpursuit,theproposedalgorithmachievesaspeed-upgainof14.4~36.7timeswhilemaintainingtheapproximationquality.
出版日期
2009年04月14日(中国期刊网平台首次上网日期,不代表论文的发表时间)