Rey-Villamizar, Nicolas

Profile Picture

Filters

20101
Reset filters

Settings

Citations

Publication Search Results

Now showing 1 - 1 of 1
  • Thumbnail Image
    Publication
    Improving nonlinear dimensionality reduction algorithms for hyperspectral data
    (2010) Rey-Villamizar, Nicolas; Rey-Villamizar, Nicolas; Vélez-Reyes, Miguel; Vélez-Reyes, Miguel; College of Engineering; College of Engineering; Manian, Vidya; Hunt, Shawn; Manian, Vidya; Hunt, Shawn; Department of Electrical and Computer Engineering; Department of Electrical and Computer Engineering; Vélez-Reyes, Miguel; Vélez-Reyes, Miguel
    Dimensionality reduction is a key step in hyperspectral image processing due to the large amount of data. Linear and nonlinear approaches have been proposed. The most important parameters on the majority of nonlinear dimensionality reduction algorithms (NLDR) are the number of neighbors used to construct the starting graph, and the number of dimensions of the low dimensional space where the data is embedded. This research work focuses on the influence of the first parameter on the DR. Newly proposed methods for constructing the weighted graph are used: k-VC, k-EC and k-MST which are alternatives to the classical approaches to k nearest neighbors (k-NN) and epsilon neighborhood (e-NN). This methods have the advantage that connectedness of the graph is guarantee and also update of the graph in case new data is available is computationally inexpensive compare to recalculating all the graph again as is needed on the classical algorithms. Also, a newly proposed neighborhood selection technique called cam-weighted neighborhood is used in combination with the NLDR algorithms. Finally, a method to improve the geodesic dis- tance estimation is explored. Experiments are carry out over hyperspectral datasets, where classification is used as the validation criteria.