Developing Metrics for Learning the Geometric Structure of Appearance Manifolds

R.C. Hoover (USA)


Manifold learning, appearance manifolds, curvature.


The current paper develops methods to analyze the local and global characteristics of one-dimensional manifolds arising in applications of pose determination in robotic vision. The approach for local analysis utilizes techniques from differential geometry to construct local coordinate frames in the high-dimensional image space. These frames provide analytical information about the local geometry of the manifold. For global analysis a distance matrix is developed to analyze how “irregular” the manifold is. Experimental results are provided by applying the proposed approach on both synthetic one-manifolds as well as one manifolds generated by real image data.

Important Links:

Go Back