步态识别与线性分类器安全研究
步态识别相关研究
学习转换矩阵算法
在步态识别中,有一个重要的算法用于学习转换矩阵,即 LearnTransformationMatrixMMC(GL) ,具体步骤如下:
Algorithm 1. LearnTransformationMatrixMMC(GL)
1: split GL = {(gn, ℓn)}NL
n=1 into identity classes {Ic}CL
c=1
such that gn, gn′ ∈Ic ⇔ℓn = ℓn′ and set pc (we set pc = Nc/NL)
2: compute overall mean μ =
1
NL
NL
n=1 gn
and individual class means μc =
1
Nc
Nc
n=1 g(c)
n
3: compute ΣB = CL
c=1 pc (μc −μ) (μc −μ)⊤
4: compute X =
1
√
NL [(g1 −μ) · · · (gNL −μ)]
5: compute Υ =
√p1 (μ1 −μ) · · · √pCL (μCL −μ)
6: compute eigenvectors Ω and corresponding eigenvalues Θ of ΣT = XX⊤through
SVD of X
7: compute eigenvectors Ξ of Θ−1/2Ω⊤ΣBΩΘ−1/2 through SVD of Θ−1/2Ω⊤Υ
8: compute eigenvectors Ψ = ΩΘ−1/2Ξ
9: c
超级会员免费看
订阅专栏 解锁全文
36

被折叠的 条评论
为什么被折叠?



