Radiometrically calibrated radiance hyperspectral images can be converted into reflectance images using atmospheric correction in order to extract useful ground information. There are some artifacts in the converted reflectance images due to spectrally misregistered sensor and atmospheric model error. These artifacts give coherent saw-tooth effects in the spectra of the reflectance imagery. These effects degrade the performance of classification and target detection algorithms and make them difficult to compare with ground target spectra. Three spectral misregistration compensation methods were developed in order to compensate for the consistent noise effects. If a ground truth spectrum exists for a test image, the ground truth spectrum can be divided by the spectrum derived from the reflectance image. This will give a coefficient indicating the difference between the ground truth spectrum and the noisy spectrum in the reflectance image. Multiplying this coefficient spectrum and the reflectance image spectrum can correct the saw-tooth effects. The other methods use the Cubic Spline smoothing technique. Cubic Spline smoothing is a fitting algorithm with a non-local smoothing method. Cubic spline smoothing can smooth out the saw-tooth noise in the spectra then the correction coefficient can be calculated as describe above. It is important to find relatively pure and unmixed pixels for the correction coefficient. Two methods for identifying relatively pure pixels were used for this research. The first is the Uniform Region method that is to identify the pixels with small standard deviation values among neighbor pixels. The second method is the Least Ratio method that is used to calculate ratios (standard deviation between smoothed and non-smoothed spectra divided by average reflectance of the spectra) and then calculate the correction coefficient using pixels having small ratios. Spectral misregistration was also simulated using MODTRAN lookup table and DIRSIG (The Digital Imaging and Remote Sensing Image Generation) synthetic image to understand and characterize the effect of spectral misregistration. The spectral misregistration compensation algorithms were tested and verified by the performance measurement of classification and target detection algorithms for test images (real and synthetic images).
Library of Congress Subject Headings
Remote sensing--Data processing--Mathematical models; Image processing--Digital techniques; Computer algorithms
Department, Program, or Center
Chester F. Carlson Center for Imaging Science (COS)
Choi, Hyeungu, "Spectral misregistration correction and simulation for hyperspectral imagery" (2002). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus