Towards automatically learning an implicit model from 2D-images based on a local similarity analysis of contours

The article deals with enabling an intelligent system to autonomously learn an implicit model of its environment. An unsupervised learning method is presented which learns the topological connections of different object views. Moreover, the method is able to distinguish between different objects. Based on a systematic local analysis of the objects' contours, the method unites learning a topology (i.e. navigation) and object recognition into one framework.