Image Processing Reference
In-Depth Information
the clustering scheme, because humans are formidable vision experts on their own.
The feature vectors discussed here are easy to interpret by a human expert, as they
encode the geometric properties. If human expertise is involved, features that are
easy to interpret by humans is preferable. In that case, it will likely be easier to use
or to devise novel features fitting to the clustering schemes that are available, than to
design novel clustering schemes for the available features. An overview of classical
clustering techniques can be found in [87, 135].
To restore the spatial continuity in Sect. 16.5, alternative approaches to the
scheme discussed above could be used. These include the use of stochastic relax-
ation and Markov random fields MRF [81].
