模式识别Chapter 2-1.ppt

《模式识别Chapter 2-1.ppt》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、1PatternRecognition北京交通大学电子信息工程学院Chapter2BayesianDecisionTheoryIntroductionBayesiandecisiontheory--error/riskminimumDiscriminantfunctionThenormaldensityDiscriminantfunctionfornormaldensityPatternRecognitionSystem4Anexample–characterRecognitionab?5Anexample–chara

2、cterRecognitionabTwoclasses,denotedbyC1,C2C1C26举例–字符识别7举例–字符识别Correspondingtoa;Correspondingtob;8举例–字符识别9Generalizationdifferentimages!正确分类的难度很大！10Features输入矢量的维数很高设法降低维数特征字符高与宽的比率v11Features字符高与宽的比率v如何选择合适的阈值？12Features特征T直方图字符高与宽的比率13Classify（分类）14Theoverallsy

3、stemcanbeviewedasamappingfromasetofinputvariables,toanoutputvariables.Theproblemishowtogetthemappingfunction.Classify（分类）15ClassifyClassifyx1x216Classify17Classify18Classify19模式识别系统的主要构成怎样才能得最好的分类效果呢？20模式识别系统的主要构成统计模式识别神经网络支持向量机……21Bayes’theoremSupposingthatwewi

4、shtoclassifyanewcharacterbutasyetwehavemadenomeasurementsontheimageofthatcharacter.Thegoalistoclassifythecharacterinsuchawayastominimizetheprobabilityofmisclassification.22Bayes’theoremIfwehavecollectedalargenumberofexamplesofthecharacter,wecouldfindthefractions

5、whichbelongineachofthetwoclasses.23Bayes’theoremWeformalizethisbyintroducingthepriorprobabilitiesofanimagebelongingtoeachclass.先验概率24Bayes’theoremPrioriprobabilities（先验概率）Iftheletter‘a’occursthreetimesasoftenastheletter‘b’25Bayes’theoremIfwewereforcestoclassifya

6、newcharacterwithoutbeingallowedtoseethecorrespondingimageThenthebestwecandoistoassignittotheclasshavinghigherpriorprobability.先验概率26Bayes’theoremThenthebestwecandoistoassignittotheclasshavinghigherpriorprobability.Thatis,weassigntheimagetoclass27Bayes’theoremNow

7、supposethatwehavemeasuredthevalueofthefeaturevariablesItisclearthatthisgiveusfurtherinformationonwhichtobaseourclassificationdecision基于特征分类28Bayes’theorem直方图29Bayes’theoremPrioriprobabilities（先验概率）30Bayes’theoremJointprobability--theprobabilitythattheimagehasthe

8、featurevalueandbelongstoclass联合概率--属于类而拥有特征值的概率31Bayes’theoremConditionalprobability--specifiestheprobabilitythattheobservationfallsincolumnofthearraygiventhatitbelon