Visual information: How to find it in images?

"A picture is worth a thousand words!" But how to find this knowledge in images? Visual scene analysis as well as medical image processing pose fundamental problems for computer science since they relate the epistemiological question "what information can be extracted from data" to the computational challenge "how efficiently can we extract relevant information from data". An information theoretic access to model validation is proposed which will address the question of structure sensitive information. Good generalization abilities of structure hypotheses are required to select models. We will demonstrate approaches to automatically extract object models from images by learning and to find discriminative features to predict survival times from tissue microarray analysis in cancer diagnosis.