ICCV13_log Euclidean kernels for sparse representation and dictionary learning

Symmetric positive definite matrices (SPD)    

Sparse Representation (SR) of SPD matrices

The space of SPD matrices -------a Lie group (Riemannian manifold)

Take full advantage of its geometric structure

A kernel-based method for SR

Dictionary Learning (DL) of SPD matrices

The space of SPD matrices, with the operations of logarithmic multiplication and scalar logarithmic multiplication defined in the Log-Euclidean framework, is a complete inner product space.

 Convariance matrices (model the second-order statistics of image features also result in SPD matrices)

在SPD稀疏编码过程中，主要考虑到的三个因素：

(1) the non-linear combination of atom matrices

(2) To evaluate the reconstruction error, the intrinsic distance between SPD matrices

(3) the updating of dictionary atoms

主要是把SR 扩展到lie group space中。