We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
Multi-Class Manifold Preserving Isomap Using Sammon's Projection.
- Authors
Mishra, Shashwati; Pradhan, Chittaranjan
- Abstract
Isometric feature Mapping (Isomap) gives promising results in preserving the original manifold structure in the case of a highly twisted and curved manifold. Classical Multi-Dimensional Scaling (MDS) uses Euclidean distance concept for obtaining low-dimensional embedding. As a distance preserving dimensionality reduction technique, Isomap gives emphasis on geodetic distances. Due to lack of any linear relationship between reduced embedded information and original high-dimensional information, it is considered a nonlinear dimensionality reduction technique. This paper concentrates on multi-class manifold geometry preservation. Like MDS, Sammon's mapping tries to preserve the manifold geometry by minimizing the Sammon's stress. Sammon's projection gives better result in preserving small distances. Sammon's algorithm was applied instead of MDS for embedding information in a lower dimension in the final step of Isomap and a more clear output was obtained.
- Subjects
ISOMETRIC projection; DATA mapping; EUCLIDEAN algorithm; GEODESY; DIMENSION reduction (Statistics); MULTIDIMENSIONAL databases; EMBEDDED computer systems
- Publication
IUP Journal of Computer Sciences, 2011, Vol 5, Issue 3, p18
- ISSN
2583-441X
- Publication type
Article