资源描述：
《The Science of Pattern Recognition：Achievements and Perspectives》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。
1、TheScienceofPatternRecognitionAchievementsandPerspectivesRobertP.W.Duin1andEl˙zbietaPekalska21ICTgroup,FacultyofElectr.Eng.,MathematicsandComputerScienceDelftUniversityofTechnology,TheNetherlandsr.duin@ieee.org2SchoolofComputerScience,UniversityofManchester,UnitedKingdompekal
2、ska@cs.man.ac.ukSummary.Automaticpatternrecognitionisusuallyconsideredasanengineeringareawhichfocussesonthedevelopmentandevaluationofsystemsthatimitateorassisthumansintheirabilityofrecognizingpatterns.Itmay,however,alsobeconsideredasasciencethatstudiesthefacultyofhumanbeings(
3、andpossiblyotherbiologicalsystems)todiscover,distinguish,characterizepatternsintheirenvironmentandaccordinglyidentifynewobservations.Theengineeringapproachtopatternrecognitionisinthisviewanattempttobuildsystemsthatsimulatethisphenomenon.Bydoingthat,scientificunderstandingisga
4、inedofwhatisneededinordertorecognizepatterns,ingeneral.自动模式识别通常被认为是这样的一个工程领域：专注于开发和评价模仿或辅助人类识别模式能力的系统，但是也可能被认为是这样的一门科学：学习人类(或其它生物系统)在所处环境中发现、区别和找出特征从而标识出观察结果的本领。模式识别中工程的观点是试图建立模拟生物识别能力的系统，通过工程中的实践，总的来说，科学上的理解在模式识别中的技术需求方面得到了发展。Likeinanyscienceunderstandingcanbebuiltfromdiffer
5、ent,sometimesevenoppositeviewpoints.Wewillthereforeintroducethemainapproachestothescienceofpatternrecognitionastwodichotomiesofcomplementaryscenarios.Theygiverisetofourdifferentschools,roughlydefinedunderthetermsofexpertsystems,neuralnetworks,structuralpatternrecognitionandst
6、atisticalpatternrecognition.象任何科学一样，对模式识别的理解能够从不同方向来建立，有时甚至是相反的观点。我们将介绍模式识别科学中的主要方法，即两种不同方向且各有两个不同种类的技术，这些技术产生了四个不同学派，粗略地可以定义为：专家系统，神经网络，结构模式识别和统计模式识别。Wewillbrieflydescribewhathasbeenachievedbytheseschools,whatiscommonandwhatisspecific,whichlimitationsareencounteredandwhichpe
7、rspectivesariseforthefuture.Finally,wewillfocusonthechallengesfacingpatternrecognitioninthedecenniatocome.Theymainlydealwithweakerassumptionsofthemodelstomakethecorrespondingproceduresforlearningandrecognitionwiderapplicable.Inaddition,newformalismsneedtobedeveloped.我们将简要地描述这
8、四个学派的发展成果，它们之间的相同点及不同点，它们各自碰到的局限性及未来发展的展望。最后，我们再来看模式识别在未来几十年所面临的挑战,这
