Fast incremental learning method based on quasi-Haar and AdaBoost classifier

Fast incremental learning method based on quasi-Haar and AdaBoost classifier

  • CN 101,937,510 A
  • Filed: 09/14/2010
  • Published: 01/05/2011
  • Est. Priority Date: 09/14/2010
  • Status: Active Application
First Claim
Patent Images

1. fast Incremental Learning method based on class Haar and AdaBoost sorter, it is characterized in that:

  • described fast Incremental Learning method is for the incorrect training sample set of discerning that increases newly, adopting the resulting crucial class Haar feature set of original training sample collection is the basis, finish the incremental learning of learning machine by increasing new crucial class Haar feature, concrete steps are as follows;

    If the incorrect recognition training sample set that increases newly is a Δ

    , the sample size of sample set Δ

    is m, sample set Δ and

    original training sample union of sets collection are Ω

    , and the total sample number of sample union Ω

    is n, and the crucial class Haar feature set that set obtains based on original training sample is Γ

    ;

    The initial weight of sample set Δ

    is u 1(i)=and 1/m, i=1 wherein, 2 ..., m;

    The initial weight of sample union Ω

    is v 1(i '"'"')=1/ (n), i '"'"'=1,2 wherein ..., n;

    M, n are natural number;

    The weights of A, normalization sample union Ω

    ;

    vt(i&

    prime;

    )
    =vt(i&

    prime;

    )
    /&

    Sigma;

    t=1n
    vt(i&

    prime;

    )
    ;

    B, seek key feature based on the sample set Δ

    ;

    1) weights of normalization sample set Δ

    ;

    Wherein t represents iterations, t=1, and 2 ..., T, T are natural number;

    2) structure Weak Classifier set on the sample set Δ

    adopts (1) formula to seek the Weak Classifier φ

    of error in classification minimum tWith crucial class Haar feature δ

    ,

View all claims
    ×
    ×

    Thank you for your feedback

    ×
    ×