Image Analysis by Means of Attribute Trees — Remote Sensing Applications

In remote sensing, many target objects have highly irregular appearance. Especially natural objects — such as clouds, isotherms and curves of terrain elevation — are typically irregular in both shape and structure. This paper explains how irregular natural objects can be modelled and processed efficiently as attribute trees. First, we present a technique for extracting image topology and shape information to attribute trees. Second, we propose techniques for analysing obtained attribute trees. The analysis involves topological descriptions (indexing) and tree matching, which are useful in pattern-recognition related applications: image retrieval, classification, tracking and monitoring. We illustrate the applicability of the proposed methods with experiments on weather radar images and auroral images.