SUBPIXEL MENSURATION OF PHOTOGRAMMETRIC TARGETS IN DIGITAL IMAGES
Abstract
Investigations into the problem of extracting high-precision geometric information from digital images are described. In particular, an image-modelling algorithm based on least squares adjustment theory is developed, with the aim of obtaining subpixel estimates of target position. The adjustment model is first derived for the one-dimensional edge-pointing problem. Tests with simulated data sets indicate that subpixel accuracies may be achieved even in the presence of considerable image noise, and that accuracy of edge location estimates is a function of spread-width and signal-to-noise ratio. An edge locator based on moment preserving gives comparable results in some instances, but in general the least squares method yields higher accuracy. An experiment involving the human measurement of edges in digital images written on film is discussed. The experiment is designed to estimate the effects of various edge characteristics on the ability to measure edge locations accurately and precisely. Results indicate that pointing accuracy is affected by the width and type of the spread-function, and by the individual observer. Precision is also affected by the spread type and by the signal-to-noise ratio. The same images are used with three digital edge operators: the least squares, moment preserving, and Hueckel algorithms. All provide subpixel accuracies, with the least squares and moment preserving methods giving comparable results. Highest accuracies are obtained where the modelled spread in the data is symmetric about the edge. The development of the least squares model for more complex targets, based on the rectangular component, is next described, and then applied to the problem of pointing to cross targets in simulated aerial imagery. The adjustments perform well generally, providing accuracies of the order of 0.05 to 0.10 pixel in target position. Numerical instabilities in the adjustment causing unsatisfactory convergence are to be examined further. This research is the first step in a combined effort utilizing photogrammetric techniques with image processing capabilities to develop the potential of digital images for photogrammetric applications and to examine the geometric effects of image processing on this potential.
Degree
Ph.D.
Subject Area
Civil engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.