Automatic classification of deformable shapes

Abstract

Let D be a dataset of smooth 3D-surfaces, partitioned into disjoint classes CL_j, j= 1, , k. We show how optimized diffeomorphic registration applied to large numbers of pairs S,S' D can provide descriptive feature vectors to implement automatic classification on D, and generate classifiers invariant by rigid motions in R^3. To enhance accuracy of automatic classification, we enrich the smallest classes CL_j by diffeomorphic interpolation of smooth surfaces between pairs S,S' CL_j. We also implement small random perturbations of surfaces S CL_j by random flows of smooth diffeomorphisms F_t:R^3 R^3. Finally, we test our automatic classification methods on a cardiology data base of discretized mitral valve surfaces.

Tasks

Reproductions