Joakim Lindblad: Precise image analysis with the Coverage model
In many aspects of digital image processing the discrete domain makes things
problematic; even such a simple thing as estimation of the perimeter of an object
becomes difficult. In this lecture I will present a framework which, based on
proper handling of partial coverage of image elements, provides high precision
results without leaving the discrete digital view. Following a basic introduction
to the approach I will give some examples of segmentation and feature extraction
methods within the framework. Finally I will present some resent results,
illustrating how the framework allows a successful combination of a discrete
representation with continuous energy minimization for high quality segmentation.