Manifold Matching using Shortest-Path Distance and Joint Neighborhood Selection

We propose a nonlinear manifold matching algorithm to match multiple data sets using shortest-path distance and joint neighborhood selection. Based on the correspondence information, a neighborhood graph is jointly constructed; then the shortest-path distance within each data set is computed from the joint neighborhood graph, followed by embedding into and matching in a common low-dimensional Euclidean space. Our approach exhibits superior and robust performance for matching data from disparate sources, compared to algorithms that do not use shortest-path distance or joint neighborhood selection.