Scientific and Technical Journal


ISSN Print 2221-3937
ISSN Online 2221-3805

The paper is devoted to a task of static images’ segmentation solution where traditional hierarchical clusterization methods and clusterization methods based on partition lose their effectiveness because of a large amount of data to be processed, their “load” with different types of disturbance and presence of complex-shaped segments. As an alternative to traditional approaches one could use clusterization methods based on data density distribution that allow to “raise” high data density distribution areas and to form arbitrary-shaped segments while suppressing abnormal outliers.

One of the most effective and popular clusterization methods based on density and designated for data processing contained in very large databases (VLDB) is DBSCAN which is based on the concepts of -reachability and -connectedness. And a distance between observations is estimated in the vector Euclid metrics. While processing images, matrix descriptions are significantly much more effective than vector descriptions. A method in a matrix version is introduced where a processing object is a two-dimensional sequence; ; ; ; wherein homogeneous segments of an arbitrary shape must be formed as a result of this task solution. It’s considered that a -value is previously undetermined, segments can intersect in such a way that each observation may belong to several clusters at the same time. As a distance estimate between two observations one can use a spherical matrix norm. A transition from the traditional vector to the matrix description simplifies significantly an algorithm numerical realization.

To provide capability for working when classes intersect, a fuzzy inference mechanism, which is a matrix modification of the well-known fuzzy -means method, is introduced to the proposed procedure.

The basic paper result is a matrix clusterizationmethod based on data density distribution contained in VLDB. The method is based on the DBSCAN modification that has to do with matrix image fragments. It estimates a membership level of every fragment to the formed clusters-segments.


1. Han J., and Kamber M., (2006), Data Mining: Concepts and Techniques, Morgan Kaufmann, San Francisco, 800 p.

2. Gan G., Ma C., and Wu J., (2007), Data Clustering: Theory, Algorithms, and Applications, SIAM, Philadelphia, 466 p.

3. Abonyi J., and Feil B., (2007) Cluster Analysis for Data Mining and System Identification, Birkhäuser, Basel, 303 p.

4. Olson D.L., and Dursun D., (2008), Advanced Data Mining Techniques, Springe, Berlin. – 180 p.

5. Xu R., and Wunsch D.C., (2008), Clustering, John Wiley&Sons, Hoboken, 358 p.

6. Szeliski R., (2011), Computer Vision. Algorithms and Applications, Springer, London, 813 p.

7.Bow S.-T., (2002), Pattern Recognition and Image Preprocessing, Marcel Dekker, Inc., New-York. – 698 p.

8. Kaufman L., and Rousseeuw P.J., (1990), Finding Groups in Data: An Introduction to Cluster Analysis, John Wiley&Sons, New-York, 342 p.

9. Duda R., and Hart P., (1976), Raspoz-navaniyeobrazov I analizscen [Pattern Recognition and Scenes Analysis], Mir, Moscow, Russian Federation, 512 p. (In Russian).

10. Forsyte D., and Pons J. Kompyuternoezreniye: sovremennyypodhod [Computer Vision: Modern Approach], (2004), Izdatelskiydom “Willians”, Moscow, Russian Federation, 928 p. (In Russian).

11 Ester M., Kriegel H.-P., Sander J., and Xu X., (1996), A Density-based Algorithm for Discovering Clusters in Large Spatial Database with Noise, Int. Conf. on Knowledge Discovery in Databases and Data Mining, AAAIO Press, Portlend, Oregon, pp. 226 – 331.

12 Bezdek J.C., and Keller J., Krisnapuram R., and Pal N.R., (2005), Fuzzy Models and Algorithms for Pattern Recognition and Image Processing, Springer Science+Business Media, Inc., New-York, 776 p.

Last download:
2017-11-19 05:59:09

[ © KarelWintersky ] [ All articles ] [ All authors ]
[ © Odessa National Polytechnic University, 2014. Any use of information from the site is possible only under the condition that the source link! ]