mf_features ( Image : : FileName : )

Creating a Markov field feature file.

The operator mf_features calculates the features necessary for the MF-image interpretation of a segmented image and stores them in a file.

The MF-based image interpretation is a general, systematic and domain independent approach for the context sensitive, purely stochastical interpretation of image regions. Hereby region related, real-valued features as well as topological relations between regions are used. These are integrated with the help of an energy function into the evaluation of the current interpretation of the regions. According to the context sensitivity, there are not only object models with one element for 1-cliques of regions, but also object models with more than one element, which model the combinations of two (2-cliques) or three (3-cliques) interpretations. With the help of a maximum-a-posteriori formulation, the result interpretation turns out to be the interpretation whose energy function has a global minimum.


Attention

Only byte images are permitted. Further, the regions of the input object should not overlap, as in general this is not reasonable for a region based image interpretation.


Parameters

Image (input_object)
image(-array) -> object : byte
An image with segmented regions.

FileName (input_control)
filename.named -> string
Name of the MF-feature file to be created (the extension .fea will be added automatically).


Example
read_image(:Reg:'fabrik':) >
gauss__(Reg:Smooth_Reg:7:) >
regiongrowing__(Smooth_Reg:Seg_Reg:1,1,5,50:) >
reduce_domain(Reg,Seg_Reg:Seg_Im::) >
mf_features(Seg_Im::'fabrik':).

Result

If the parameter values are correct, the operator mf_features returns the value TRUE. Otherwise an exception is raised.


Possible Predecessors

reduce_domain


Possible Successors

mf_db_generation, mf_read_db, mf_interprete_image


See also

mf_db_generation


References

Helmut Kristen: "Markov-Feld-basierte Bildinterpretation mit automatisch generierten Datenbasen"; Diplomarbeit; Technische Universität München, Institut für Informatik, Lehrstuhl Prof. Radig; 1991.

J.W. Modestino, J. Zhang: "A Markov Random Field Model-Based Approach to Image Interpretation"; Conference on Computer Vision and Pattern Recognition; pp. 458 - 465; IEEE, San Diego; 1989.

J. Zhang: "Two-Dimensional Stochastic Model-Based Image Analysis"; Ph.D. Thesis; Rensselär Polytechnic Institute, Troy, New York; August 1988.

S. Geman, D. Geman: "Stochastic Relaxation, Gibbs Distribution, and the Bayesian Restoration of Images"; M. A. Fischler: Readings in Computer Vision; pp. 564 - 584; Morgan Kaufmann Publishers; 1987.

E.H.L. Aarts, J. Korst: "Simulated Annealing and Boltzmann Machines"; John Wiley Sons Ltd., Chichester; 1989.



Copyright © 1996-1997 MVTec Software GmbH