Image Compression by a Time Enhanced Self Organizing Map

  • Pascual Campoy
  • Pedro Gutiérrez
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 4225)


This paper presents the promising results of an innovative modification of the Kohonen’s algorithm, the time enhanced self-organizing map (TESOM), when used for low bitrate image compression. The time enhanced map is used in this paper to learn codebooks of subimages, in a similar way as other classical algorithms based on LVQ or SOM do, but taking advantage of the fact that it learns the sequence order of the input data (i.e. subimages) during the training phase. The codebook learned by the new proposed algorithm TESOM presents the advantage that the vicinity of the codes in the output map is not only established by their visual similarity, as in SOM, but also by the sequential order of the subimages during the training phase. Since this sequential order of the subimages determines the vicinity of the codes, the increment of the representative code of two consecutive subimages has been proved to have a lover Entropy and can therefore be codified by a lower bit rate. The advantage of the proposed algorithm is thoroughly evaluated and quantified over a set of experiments, which include several images, used in different ways in the training phase for codebook design and in the compression phase, and a variety of parameters.


Time enhanced self-organizing map image compression vector quantization low bitrate entropy 


  1. 1.
    Nasrabadi, N.M., King, R.A.: Image coding using vector quantization: a review. IEEE Transactions on Communications 36(8), 957–971 (1988)CrossRefGoogle Scholar
  2. 2.
    Linde, Y., Buzo, A., Gray, R.M.: An algorithm for vector quantizer design. IEEE Transactions on Communications 28(1), 84–95 (1980)CrossRefGoogle Scholar
  3. 3.
    Kohonen, T.: The self-organizing map. Proceedings of the IEEE 78(9), 1464–1480 (1990)CrossRefGoogle Scholar
  4. 4.
    Wei, H.-C., Chang, Y.-C., Wang, J.-S.: A Kohonen-based structured codebook design for image compression. In: TENCON 1993 IEEE Region 10 Conference on Computer Communication, Control and Power Engineering, Proceedings of, vol. 3, pp. 426–429 (1993)Google Scholar
  5. 5.
    Amerijckx, C., Verleysen, M., Thissen, P., Legat, J.-D.: Image compression by selforganized kohonen map. IEEE Transactions on Neural Networks 9(3), 503–507 (2002)CrossRefGoogle Scholar
  6. 6.
    Kangas,: Phoneme recognition using time-dependent versions of selforganizing maps. In: IEEE International Conference on Acoustics, Speech and Signal Processing, Proceedings of, pp. 101–104 (1991)Google Scholar
  7. 7.
    James, L., Miikkulainen, R.: SARDNET: A self-organizing feature map for sequences. In: NIPS 1994 Advances in Neural Information Processing Systems, Proceedings of, pp. 577–584 (1994)Google Scholar
  8. 8.
    Euliano, R., Principe, J.C.: Spatiotemporal self-organizing feature maps. In: International Joint Conference on Neural Networks, Proceedings of, vol. 4, pp. 1900–1905 (1996)Google Scholar
  9. 9.
    Campoy, V.C.J.: Residual Activity in neurons allows SOMs to learn temporal order. In: International Conference on Artificial Neural Networks, Warsaw, Poland, September 11-15 (2005)Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2006

Authors and Affiliations

  • Pascual Campoy
    • 1
  • Pedro Gutiérrez
    • 1
  1. 1.Departamento de AutomáticaIngeniería Electrónica e Informática IndustrialSpain

Personalised recommendations