Improving a mesh segmentation algorithm based on non-negative matrix factorization

Jisun Kang, Purdue University

Abstract

3D Mesh segmentation is used in various applications such as object recognition, reconstruction, and analyzing structure of meshes. The method for 3D mesh segmentation based on sparse non-negative matrix factorization (NMF) was previously proposed. It represents a novel, and conceptually simpler, method than other comparable algorithms. However, this method still has potential to improve performance, results could have better consistency and uniqueness with faster computation time than the prior proposed algorithm. This study introduced several approaches to enhance the performance of the algorithm comprehensively: applying dierent update rule and initialization of factor matrices, and imposing sparseness to the factor matrices and the distance matrix. In addition, we introduced how to measure the performances of the results.

Degree

M.S.

Advisors

McGraw, Purdue University.

Subject Area

Computer science

COinS