特征值正交分解 A=PDP', where P'=inverse(P), D is diagonal matrix of eigenvalues.特点: A^{n} = PD^{n}P'; e^{A}=Pe^{D}P'
QR分解 A=QR, where R is an upper triangular matrix and Q is an orthogonal matrix. QR 分解常用于解线性方程组。
LU分解 A=LU, where L is an lower triangular matrix and U an upper triangular matrix . QR 分解常用于解Ax=b=>L(Ux)=b=>Ly=b, Ux=y。
SVD分解 A=UDV', where A is m-n matrix, and D is n-n diagonal matrix. U is m-n matrix and V is n-n matrix such that U'U=V'V=I. 不是所有的矩阵都可以特征值分解,但是所有的实矩阵都可以特征值分解。
SVD一个用途是二次误差最佳逼近。这个含义上SVD, PCA, Fourier很像。
在mutiview里面,SVD几乎是处理fundmental matrix的必用工具。
further reading:
- Direct Least Squares Fitting of Ellipses by Fitzgibbon
- http://www.uwlax.edu/faculty/will/svd/index.html
- Invariant Subspaces and Condition and Singular Value Decomposition
- A few properties of Singular Value Decomposition