摘要
ConventionalGaborrepresentationanditsextractedfeaturesoftenyieldafairlypoorperformanceinextractingtheinvariancefeaturesofobjects.Toaddressthisissue,aglobalGaborrepresentationmethodforraisedcharacterspressedonlabelisproposedinthispaper,wheretherepresentationonlyrequiresfewsummationsontheconventionalGaborfilterresponses.Featuresarethenextractedfromthesenewrepresentationstoconstructtheinvariantfeatures.ExperimentalresultsclearlyshowthattheobtainedglobalGaborfeaturesprovidegoodperformanceinrotation,translation,andscaleinvariance.Also,theyareinsensitivetoilluminationconditionsandnoisechanges.ItisprovedthatGaborfilterscanbereliablyusedinlow-levelfeatureextractioninimageprocessingandtheglobalGaborfeaturescanbeusedtoconstructrobustinvariantrecognitionsystem.
出版日期
2008年03月13日(中国期刊网平台首次上网日期,不代表论文的发表时间)