SPIE Student Chapter
Seminar Series
hosted by Department of Biomedical Engineering
“Spectral Remote Sensing: What is the "dimension" of my image, how do I calculate it, and why do I care?”
David W. Messinger, Ph.D.
Director, Digital Imaging and Remote Sensing Laboratory
Chester F. Carlson Center for Imaging Science
Rochester Institute of Technology
Tuesday, 9 June 2009
1:00-2:30 PM
Room 109, Goergen Hall
University of Rochester
Abstract
Spectral remote sensing uses advanced digital imaging techniques to collect images, typically from aircraft or satellites, in not one spectral band (i.e., black and white), or three bands (i.e., blue, green, and red), or even the rainbow of seven colors (i.e., ROYGBIV). Instead, we are able to collect hundreds of spectral bands for each pixel on the ground typically covering wavelengths from 0.4 - 2.0 um. This allows us to use techniques from spectroscopy to better differentiate between materials on the ground as each unique material has a unique spectral reflectance. Typical applications include land cover classification and target detection. In some sense, these images are measured in hundreds of "dimensions" - one for each spectral
band collected. However, due to the physics of the photon interactions along the sun-surface-sensor path, the data can be correlated and don't "fill up" the full hyperspace. A method will be presented using a simple algorithm based on Point Density Estimation to calculate the inherent dimensionality of samples from a spectral image and we will show that it is typically very small (~5-10). We will also show how the dimension estimation methodology can lead to algorithms to extract information out of the image.
No comments:
Post a Comment