Projection pursuit and a VR environment for visualization of remotely sensed data

• Main Authors: Petrakos, M., Dicarlo, W., Kanellopoulos, I.
• Format: Conference Proceeding
• Language: English
• Subjects: Covariance matrix; Data visualization; Extraterrestrial measurements; Humans; Microwave integrated circuits; Pursuit algorithms; Remote sensing; Satellites; Statistical analysis; Virtual reality
• DOI: 10.1109/IGARSS.1999.771555

Humans have an exceptional capacity to identify patterns, clusters and a variety of characteristics in 1- and 2-dimensional spaces. In recent years immersive techniques and virtual reality have provided a third dimension. However, we are usually intimidated if we go up to 4 dimensions and we cannot identify but the simpler patterns in 5 or 6-dimensional spaces (with the use of glyphs). Fuzzy data contributes to further complication and each pixel is not only associated with several spectral values but with several class membership values as well. Projection pursuit is a statistical method that seeks interesting projections of a high dimensional space into a lower one according to a measure of interestingness (projection index). Projection pursuit provides therefore a flexible and convenient framework that can include current approaches to the dimensionality reduction problem (principal components, discriminant analysis) as well as fresh approaches. Two-dimensional and three-dimensional projection pursuit algorithms have been developed for the visualization of multispectral data (7 band Landsat-TM) with projection indices, which maximize class separability. An immersive environment with a data-brushing tool has been incorporated for the visualization of the three-dimensional projection pursuit output. Mixtures of RGB triplets represent fuzzy memberships.