bundle min hash

• Visual words and bundling

  it was proposed to bundle multiple SIFT features that lie in the same MSER region into a single description [26].

• 用SIFT获取所有特征点

todo

• 找center point的邻居节点

  因为是128维的高维向量，所以这里用到了kd-tree，来搜寻邻居特征点

  用到的kdtree的库，flann kdtree

• mini-hash

  The min-Hashes can be computed by generating multiple random permutations on the range of all visual words and storing them in a lookup table.