AdaBoost算法及应用.ppt

?Overview?The AdaBoost Algorithm?How and why AdaBoost works??AdaBoost for Face DetectionOverviewAdaBoostAdaptiveA learning algorithmBuilding a strong classifier a lot of weaker onesBoosting1{ 1 }), 1(h x???...weak classifiersslightly better than random1( )( )TttTth xH xsign??? ??? ?? ??2{ 1 }), 1(h x???{ 1(}), 1Th x???strong classifier1{ 1 }), 1(h x???...weak classifiersslightly better than random1( )( )TttTth xH xsign??? ??? ?? ??2{ 1 }), 1(h x???{ 1(}), 1Th x???strong classifier训练一个弱分类器(特征f)就是在当前权重分布的情况下,确定f 的最优阈值以及不等号的方向,使得这个弱分类器(特征f)对所有训练样本的分类误差最低。1 ( )( , , , )0pf x ph x f p????????? ???????????????????其他1{ 1 }), 1(h x???...weak classifiersslightly better than random1( )( )TttTth xH xsign??? ??? ?? ??2{ 1 }), 1(h x???{ 1(}), 1Th x???strong classifierHow good the strong one will be?How good the strong one will be?The AdaBoost AlgorithmGiven:1 1 where ( , ), , ( , ) , { 1, 1}m m i ix y x y x X y? ????Initialization: 11( ) , 1, ,mD i i m? ??For :1, ,t T??? Find classifier which minimizes error wrt Dt ,.,: { 1, 1}th X???1 where argmin ( )[ ( )]jmt j j t i j iihh D i y h x? ??? ???:probability distribution of 's at time ( )t iD i x t? Weight classifier:11ln2ttt?????? Update distribution:1( ) exp[ ( )], is for normalizati( )ont t i t it ttD i y h xD i ZZ????minimize weighted errorfor minimize exponential lossGive error classified patterns more chance for :1 1 where ( , ), , ( , ) , { 1, 1}m m i ix y x y x X y? ????Initialization: 11( ) , 1, ,mD i i m