Paper: Combing multiple manifold-valued descriptors for improved object recognition (Mehrtasg Harandi)
A learning method for classification using multiple manifold-valued features
1、Feature combination: Combing HOG and Region Convariance descriptors that reside on two different manifolds.
2、A kernel on the n-dimensional unit sphere
一些非向量空间的描述子有:
Normalized histogram vectors、Convariance descriptors (object recognition)
Diffusion tensors (biomedical image analysis)
3D rotation matrices (geometrical computer vision)
Linear subspaces of the n-dimensional Euclidean space (video vision)
分析这种非向量空间描述子的方法是利用黎曼几何~
把欧式算法推广到黎曼流形的方法:
一是,obtain a Euclidean represenration of the manifold-valued data by approximating the manifold by the tangent space. (log and exp maps)
二是,embed the manifold in the high dimensional Reproducing Kernel Hilbert space (RKHS) using a positive definite kernel.
Combining multiple manifold-valued descriptor via RKHS embedding.