Speakers Bio & Abstract

Abstract Air- and spaceborne acquisition of cartographic informationImage analysis can be defined as the automatic derivation of an explicit and meaningful description of the object scene depicted in the images. For this purpose, individual objects must be recognized and described. This recognition needs prior knowledge of objects in terms of models, which must be made available to the machine. Alternatively, they can also be learnt in a first step of the process itself. The set up of the object models is a major problem of image analysis. At present, it is still not clear, which elements of an object and scene description need to be taken into account. Recently, more and more statistical methods are used in knowledge acquisition and representation. Presently, these attempts are still provisional, however it is obvious that an efficient automatic generation of models is a decisive prerequisite for the success of image analysis altogether. Another possibility for introducing a priori knowledge is based on the assumption that images are normally analysed for a certain purpose, pre-defined at least in its main features. In GIS, for example, the available information is described in object catalogues, which contain relevant information for formulating the object models for image analysis. Available GIS data may also be used as part of the knowledge base. In this talk, I will give an overview of the research in photogrammetry and remote sensing, and in image analysis in particular, our group in Hannover has conducted over the last years. Examples will comprise work on road extraction from high-resolution images using snakes, level sets and normalised cut segmentation, multi-temporal classification of satellite images based on conditional random fields, automatic quality control of topographic data using aerial and satellite images, and the integration of vector and height data to generate a consistent 3D topographic database. Through these examples it will become clear how much of the vision of automatic image analysis has been achieved today, and which questions still await a solution.