Pattern recognition and image analysis are impicitly concerned with objects represented by sets of pixels in digitized images. In pattern recognition the final goal is to recognize the object in the image under consideration. Conversely in image analysis the object in an image is primarily mapped to a set of quantitative features. These features or measurements combined with external properties allow subsequent interpretation of the image objects.
In pattern recognition and image analysis the most cumbersome processing step is the definition, segmentation or delineation of the actual object inside of the image. Numerous publications exist describing methods and algorithms performing the task [11][4]. Most of the proposed algorithms contain explicit and implicit parameters adjusted to the problem.
Using concepts from mathematical topology a method is described applying only procedural declarations but no a-priori knowledge like numeric parameters. The declarations are applied to devide an arbitrary digital grey value image into different objects. Additionally these objects are related to each other which is one condition for further evaluation of image content.
The methods used base on minmax operations, mentioned in [4] and more recently by [13]. Relations exist to the concept of medial axis, see [1]. The algorithms used are already published in [8][5]. In this article the results and the possible applications of the transformations are outlined in more detail.