Feature detection in grayscale aerial images
Katherine Treash · DSpace@MIT (Massachusetts Institute of Technology) · 1999
Feature detection in aerial images entails a number of specific problems, depending on the feature to be detected as well as the method to be used. This thesis focuses on the problem of automatically detecting roads in grayscale aerial images. The challenges of this particular problem are discussed and two systems are proposed as solutions. Both systems are edge-based methods and have two key steps: an edge detection step followed by an edge linking procedure. One system uses a variant of the Nevatia-Babu edge detector for the first step. This edge detection method revolves around convolutions of the aerial image with a series of masks and is quite simple to implement. The other system applies a wavelet edge detector to the images. Wavelets are briefly introduced and then a set of wavelet filters, developed by Mallat and Zhong, are detailed. In both road detection systems, the edge linking technique is the same. A new edge linking algorithm is developed using a zone-based technique, and is designed to link the long, low-curvature edges which represent roads. The results of applying each edge detection technique individually, followed by the edge linking procedure, to three test images are displayed. These results are compared and contrasted to try to