An automated approach to change detection analysis was developed for use in multi-temporal image comparisons. An algorithm was developed which enables the user to perform automatic image-to-image rectification. Manual registration techniques are utilized to register a reference image to a Universe Transverse Mercator map projection. Control points, or kernel images, are extracted from the rectified reference image and located automatically in the to-be-mapped images via mathematical correlation. A two windowed approach is used that requires an estimation of the location of the control point in the to-be-mapped image. This estimate is used to create a search area which is correlated with the kernel image. The images were rectified to within approximately two pixels. The images were radiometrically normalized so that actual ground changes can be distinguished from those that occur due to imaging conditions. This was done through a simple histogram matching technique. Next, the images were classified to illustrate the changes in land cover type. An unsupervised classification was used to train the reference image. The rest of the images in the set were classified using the spectral signature data generated from this process. The classification accuracy was dependent on the normalization procedure used. The process was demonstrated using LANDSAT MSS imagery to show the extent to which the logging technique of clearcutting has devastated the forest stands in the North Cascades of Washington state.
Library of Congress Subject Headings
Forests and forestry--Remote sensing; Remote sensing--Data processing; Forest mapping
Department, Program, or Center
Chester F. Carlson Center for Imaging Science (COS)
Moriarty, Kaleen S., "Automated image-to-image rectification for use in change detection analysis as applied to forest clearcut mapping" (1993). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus