Extended .Isornap for Classification
Ming–Hsuan Yang · 2002
The Isomap method has demonstrated promising msulrs in finding a low dimensional embedding from samples in the high dimensional input space. The crux of this merhod is ro estimate geodesic distance with multidimensional scaling for dimensionaliry reduction. Since the Isomap merhod is developed based on rhe reconsrruction principle, it may not be optimal fmm the classification viewpoint. We present an extended lsomap method thar utilizes Fisher Linear Discriminant for pattern classijication. Numerous experiments on image data sers show that our extension is more effective than the original lsomap method for pattern classifrcarion. Furthermore, the exrended lsomap shows promising results compared with best classijication methods in the literature.