Change Detection and Analysis in Multi-Spectral Images,

摘要

This thesis describes research toward the development of a general image understanding system. The system has been directed toward the problem of the comparison of pairs of different images of the same scene to generate descriptions of the changes in the scene. Unlike earlier work in the change analysis area, all the matching and change analysis were performed at a symbolic level rather than a signal level. To facilitate this symbolic analysis over a wide variety of images, advances in several other areas of image analysis were also required. Several diverse scenes (house, cityspace, satellite images, aerial images, and radar images) are presented, each of which has a task description and a predefined set of knowledge elements, and will show how several different tasks can be performed with a general change analysis system. Early segmentation techniques were either designed for specific applications or were very expensive. The segmentation of an image into regions by a histogram based region splitting procedure has proved to be useful over a wide range of images, but also tends to be expensive (Ohlander, 1975). In order to make this procedure more efficient, the use of 'planning' was incorporated into the segmentation processing. This use of planning means that the segmentation is generated in about a tenth (or better) of the time required without planning.