The main objective of any imaging system is to collect information. Information is conveyed in remotely sensed imagery by the spatial and spectral distribution of the energy reflected/emitted from the earth. This energy is subsequently captured by an overhead imaging system. Post-processing algorithms, which rely on this spectral and spatial energy distribution, allow us to extract useful information from the collected data. Typically, spectral processing algorithms include such procedures as target detection, thematic mapping and spectral pixel unmixing. The final spectral products from these algorithms include detection maps, classification maps and endmember fraction maps. The spatial resolution, spectral sampling and signal-to-noise characteristics of a spectral imaging system share a strong relationship with one another based on the law of conservation of energy. If any one of these initial image collection parameters were changed then we would expect the accuracy of the information derived from the spectral processing algorithms to also change. The goal of this thesis study was to investigate the accuracy and effectiveness of spectral processing algorithms under different image levels of spectral resolution, spatial resolution and noise. In order to fulfill this goal a tool was developed that degrades hyperspectral images spatially, spectrally and by adding spectrally correlated noise. These degraded images were then subjected to several spectral processing algorithms. The information utility and error characterization of these "degraded" spectral products is assessed using algorithm-specific metrics. By adopting a factorial designed experimental approach, the joint effects of spatial resolution, spectral sampling and signal-to-noise with respect to algorithm performance was also studied. Finally, a quantitative performance comparison of the tested spectral processing algorithms was made.
Library of Congress Subject Headings
Image processing--Mathematical models; Algorithms; Imaging systems; Image processing--Digital techniques
Department, Program, or Center
Chester F. Carlson Center for Imaging Science (COS)
Klatt, John, "Error characterization of spectral products using a factorial designed experiment" (2001). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus