Analysis and Development of Image Statistics and Redundancy Removal.

Abstract : The goal of classifying objects of cartographic interest in aerial photographs was approached using techniques from pattern recognition and image processing. Bridge and airport images were chosen as the initial objects of interest and segments of photographs containing them were digitized for the data base. Edge-detection and Hough transform algorithms identified structures as candidate bridges; additional decision logic (using global contrast and other attributes) further reduced the set. Results indicate the feasibility and low computational cost of the approach. Additional results in discrete medial-axis transformation are presented, as are methods for encoding the two kinds of images. The characteristics of the two kinds of targets are so distinctive that encoding promises substantial efficiencies. (Author)