Development of region based image analysis techniques

Abstract

Region based image analysis is the one of the important techniques of the image understanding system. In this thesis, the fundamental techniques which are useful in the region based image analysis are developed. A contour matching technique based on the residual error of the least squares fit is described and the matching of the flying aircrafts is demonstrated. The performances are compared with the previous matching techniques. To obtain the perfect image descriptions the concept of the super regions is considered. The super region is defined as the region which consists of multiple regions which are connected together and must be useful in the region based image analysis. An algorithm of the extraction of the super region in the region based image which is based on the region adjacency matrix is proposed, and three examples demonstrating the usefulness of the algorithm are given. Finally, among many regions a selection method of the significant regions which can be effectively used in the region based image matching is proposed, using the overlapping ratio between the edge points and the boundary points of the region and three examples are given, displaying the reasonable results of the proposed method. And to demonstrate how to apply the techniques developed in this thesis to the region based image matching, a suggestion of the region based image matching is made.