Advertisement

A structured neural network invariant to cyclic shifts and rotations

  • Sabine Kröner
Object Recognition
Part of the Lecture Notes in Computer Science book series (LNCS, volume 1296)

Abstract

Shift and rotation invariant pattern recognition is usually performed by first extracting invariant features from the images and second classifying them. This poses the problem of not only finding suitable features but also a suitable classifier.

Here a class of structured invariant neural network architectures (SINN) is presented that performs adaptive invariant feature extraction and classification simultaneously. The special characteristic of the pyramidal feedforward architecture of the SINN is sparse connectivity and the use of shared weight vectors. This guarantees the invariance of the network output with respect to cyclic shifts and rotations of the input image. In experiments the recognition ability of the SINN is shown on a database of textile images. Without any preprocessing of the images and without the need to choose an appropriate classifier the SINN achieves similar or even better results than standard pattern recognition methods.

Keywords

Recognition Rate Invariant Feature Feature Extraction Method Cyclic Shift Invariant Integration 
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.

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

References

  1. 1.
    E. Barnard, D. Casasent, Shift Invariance and the Neocognitron, in Neural Networks, Band 3, S. 403–410, Pergamon Press, 1990Google Scholar
  2. 2.
    K. Fukushima, N. Wake, Handwritten Alphanumeric Character Recognition by the Neocognitron, IEEE Trans. on Neural Networks, Vol. 2, No. 3, May 1991Google Scholar
  3. 3.
    T. Kanaoka et al., A higher-order neural network for distortion invariant pattern recognition, Pattern Recognition Letters, Vol. 13, pp. 837–841, Dec. 1992CrossRefGoogle Scholar
  4. 4.
    S. Kröner, A Neural Network for Calculating Adaptive Shift and Rotation Invariant Image Features, Proc. of EUSIPCO'96, 8th European Signal Proc. Conf., G. Ramponi et al. (Eds.), Vol. II, pp. 863–866, Trieste, Italy, Sept. 1996Google Scholar
  5. 5.
    S. Kröner, A Learning Algorithm for Structured Invariant Neural Networks, Technical Report 5/96, Technische Informatik I, TU Hamburg-Harburg, Sept. 1996.Google Scholar
  6. 6.
    S. Kröner, H. Schulz-Mirbach, Fast adaptive calculation of invariant features, Tagungsband Mustererkennung 1995, (17. DAGM Symposium), Sagerer, G. et al. (Eds.), S. 23–35, Bielefeld, Sept. 1995. Reihe Informatik aktuell, Springer VerlagGoogle Scholar
  7. 7.
    C. Li, and C.-H. Wu, Introducing rotation invariance into the Neocognitron model for target recognition, Pattern Recognition Letters, Vol. 14, pp. 985–995, Dec. 1993CrossRefGoogle Scholar
  8. 8.
    Y. le Cun: Generalization and Network Design Strategies, Connectionism in Perspective, R. Pfeiffer, Z. Schreter, F. Fogelman-Soulié, L. Steels (Eds.), Elsevier Science Publishers B.V. 143–155, North-Holland, 1989Google Scholar
  9. 9.
    D. de Ridder: Shared weight neural networks in image analysis M.Sc. Thesis, TU Delft, March 1996Google Scholar
  10. 10.
    D. Rumelhart, J. McClelland, (Ed.), Parallel Distributed Processing, Vol. 1, MIT Press, Cambridge, Massachusetts, 1986Google Scholar
  11. 11.
    H. Schulz-Mirbach, Ein Referenzdatensatz zur Evaluierung von Sichtprüfungsverfahren für Textiloberflächen, Tech. Report 4/96, Tech. Inf. I, TU Hamburg-Harburg, Sept. 1996Google Scholar
  12. 12.
    H. Schulz-Mirbach, Anwendung von Invarianzprinzipien zur Merkmalgewinnung in der Mustererkennung, VDI-Fortschrittbericht, Reihe 10, Nr. 372, 1995Google Scholar
  13. 13.
    L. Spirkovska, M.B. Reid, Coarse-Coded Higher-Order Neural Networks for PSRI Object Recognition, IEEE Trans. Neural Networks, Vol. 4, No. 2, 1993Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 1997

Authors and Affiliations

  • Sabine Kröner
    • 1
  1. 1.Technische Informatik ITechnische Universität Hamburg-HarburgHamburg

Personalised recommendations