Manifold Alignment via Local Tangent Space Alignment
Gelan Yang, Xue Jun Xu, Jianming Zhang · 2008
Manifold alignment (Ham et al., 2005) is about mapping several datasets into a global space, and is of great importance in learning the shared latent structure (Shon et al., 2006), data fusion and multicue data matching (Lafon et al., 2006). In this paper, we propose an algorithm to solve this problem via local tangent space alignment (Zhang et al., 2004) (LTSA). LTSA is used here as a method to find the inner manifold constraint of each dataset. A cost function to measure the quality of alignment is given by combining the inner manifold constraints of each dataset and the matching points constraints among different datasets. The effectiveness of our algorithm is validated by applying it to the problem of image sequences alignment.