PRECISION FEATURE LOCATION IN DIGITIZED IMAGERY (THRESHOLD DECOMPOSITION, SUBPIXEL)
Abstract
Methods for the precision location of features in digitized imagery are developed. These features include one- and two-dimensional grey level step edges and two-dimensional objects. A new moment based edge locator is developed which is able to determine digitized step edges with subpixel accuracy. Under appropriate assumptions, these step edges can be located with an accuracy limited only by the signal-to-noise ratio and not by the pixel size. In addition, this operator can be used as an edge detector with excellent results when compared with other edge detectors. Another method involves using threshold decomposition on a grey level object. The summation of many contour locations from a single object results in accurate subpixel location. Theoretical work shows that the location error arises from two-dimensional corners. Maximum location error bounds are determined for squares, rectangles, and crosses. Maximum location error ranges from 0.06 to 0.10 pixels. Average location error results show that accuracies are in the neighborhood of 0.02 pixels. An implementation of the threshold decomposition method is developed and applied to crosses in aerial imagery. Results from this implementation verify the theoretical maximum location error bounds. Application of the moment based edge operator to determine edge locations in stereo images is developed. Results show that as the resolution in the stereo images is increased by the subpixel accuracy of the edge operator, a greater resolution in depth is obtained. Empirical results from real digitized imagery show gains of 4 to 5 times the original depth resolution.
Degree
Ph.D.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.