A local diffusion wavelet approach for scattered data registration based on GPU

In order to register partly overlapped data clouds measured from different view points, a multi-resolution registration method based on local diffusion wavelet features was proposed. Firstly, we detect multi-scale salient points from large number of original data, and a subset of data was extracted within a neighbourhood around each salient point. Secondly, diffusion wavelet decomposition was performed on each neighbourhood data set. We constructed a multi-resolution feature descriptor using wavelets coefficients for each of feature points. The correspondences were computed between two sets of feature points by measuring the similarity of feature descriptors. The algorithms were implemented in a parallel way on GPU to improve the efficiency.