PDF version of this document
The concept above has been applied to select good descriptors of shape [42]. Selection of points that describe a given shape, as explained in Section 2, was perpetually altered and evaluated to find shape models and examples that require a smaller set of data to be passed as an encoded message.
To express the process at a moderate pace, each time points on the curve that trace the shape are selected, a different model is ultimately constructed. A good and compact statistical model is one whose legal variations are relatively small and possibly so are the number of its control points. Such a model is sought via a general optimisation regime under which point are reparameterised. MDL can be used in an objective function that is iteratively evaluated for each such points reparameterisation. The minimisation process was described in reasonable detail in Subsection 3.7 on optimisation. The more genuine part of this work is the use of an existing information theoretic measure, namely MDL, to guide an autonomous search for good models. This work will be explained with respect to current research in the next section.