Learning Translation Invariant Shape Knowledge for Steering Diffusion-Snakes
Daniel Cremers, Christoph Schnörr, Joachim Weickert, Christian Schellewald · 2000
. Biological vision is characterized by an intricate interplay of external sensory input and previously acquired internal representations of the world. In this work, we present a computer vision model for image segmentation which allows to combine external visual input and internally represented shape knowledge in a unifying energy functional. We explain how prior shape information is acquired. Numerical examples show how the relative weight of prior information affects the outcome of segmentation. A method to include translation invariance clarifies the way in which shape knowledge is acquired. The behavior of our method in a real-world scenario is demonstrated.