Image Classification by Histogram Features Created with Learning Vector Quantization

  • Marcin Blachnik
  • Jorma Laaksonen
Part of the Lecture Notes in Computer Science book series (LNCS, volume 5163)


Histogram-based data analysis is one of the most popular solutions for many problems related to image processing such as object recognition and classification. In general a histogram preserves more information from the first-order statistics of the original data than simple averaging of the raw data values. In the simplest case, a histogram model can be specified for a specific image feature type independently of any real image content. In the opposite extreme, the histograms can be made dependent not only the actual image contents, but also on the known semantic classes of the images. In this paper, we propose to use the Learning Vector Quantization (LVQ) algorithm in fine-tuning the codebook vectors for more efficient histogram creation. The performed experiments show that the accuracy of the Interest Point Local Descriptors (IPLD) feature for image classification can be improved by the proposed technique.


Support Vector Machine Receiver Operator Characteristic Curve Area Under Curve Interest Point Learn Vector Quantization 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 1.
    ISO/IEC: Information technology - Multimedia content description interface - Part 3: Visual, 15938-3:2002(E) (2002)Google Scholar
  2. 2.
    Lowe, D.G.: Distinctive image features from scale-invariant keypoints. International Journal of Computer Vision 60(2), 91–110 (2004)CrossRefGoogle Scholar
  3. 3.
    Dorko, G., Schmid, C.: Selection of scale-invariant parts for object class recognition. In: Proceedings of Ninth IEEE International Conference on Computer Vision, vol. 1, p. 634. IEEE Computer Society, Los Alamitos (2003)CrossRefGoogle Scholar
  4. 4.
    Kohonen, T.: Self-Organizing Maps. Springer, Heidelberg (2001)zbMATHGoogle Scholar
  5. 5.
    Kuncheva, L., Bezdek, J.: Presupervised and postsupervised prototype classifier design. IEEE Transactions on Neural Networks 10(5), 1142–1152 (1999)CrossRefGoogle Scholar
  6. 6.
    Blachnik, M., Duch, W., Wieczorek, T.: Selection of prototypes rules: context searching via clustering. In: Rutkowski, L., Tadeusiewicz, R., Zadeh, L.A., Żurada, J.M. (eds.) ICAISC 2006. LNCS (LNAI), vol. 4029, pp. 573–582. Springer, Heidelberg (2006)CrossRefGoogle Scholar
  7. 7.
    Kuncheva, L., Bezdek, J.: Nearest prototype classification: Clustering, genetic algorithms or random search? IEEE Transactions on Systems, Man, and Cybernetics C28(1), 160–164 (1998)Google Scholar
  8. 8.
    Hoppner, F., Klawonn, F., Kruse, R., Runkler, T.: Fuzzy Cluster Analysis. Wiley, Chichester (1999)Google Scholar
  9. 9.
    Kuncheva, L., Bezdek, J.: An integrated framework for generalized nearest prototype classifier design. International Journal of Uncertainty 6(5), 437–457 (1998)zbMATHGoogle Scholar
  10. 10.
    Everingham, M., Van Gool, L., Williams, C.K.I., Winn, J., Zisserman, A.: The PASCAL Visual Object Classes Challenge 2007 (VOC2007) Results (2007),
  11. 11.
    Joachims, T.: Learning to Classify Text Using Support Vector Machines. Kluwer Academic Publishers, Dordrecht (2002)Google Scholar
  12. 12.
    Bermejo, S.: Learning with nearest neighbor classifiers. PhD thesis, Technical University of Catalonia (2000)Google Scholar
  13. 13.
    Tipping, M.: The relevance vector machine. In: Advances in Neural Information Processing Systems. Morgan Kaufmann, San Francisco (2000)Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2008

Authors and Affiliations

  • Marcin Blachnik
    • 1
    • 2
  • Jorma Laaksonen
    • 2
  1. 1.Electrotechnology DepartmentSilesian University of TechnologyPoland
  2. 2.Adaptive Informatics Research CentreHelsinki University of TechnologyTKKFinland

Personalised recommendations