Invariance via group-integration: a feature framework for 3D biomedical image analysis
Janis Fehr, Olaf Ronneberger, Janina Schulz, Thorsten Schmidt, Marco Reisert, H. J. Burkhardt · 2008
One very generic approach towards the construction of features achieves invariance against a certain transformation via integration over the respective mathematical group. In this paper we present a general framework for invariant feature design via group integration for biomedical image analysis on 3D volumetric data, and show the common mathematical context of several previously published invariant methods which are all covered by this basic framework. We focus on the mathematical design paradigms of such features and provide fast implementation methods. Further we embed a priori knowledge into the design of highly specialized features. Practical applications to these methods are shown on several different datasets with a wide range of different biomedical image analysis problems.