Abstract
This paper presents a rationale for the design of optimal sequential sampling procedures for multi-hypothesis discrimination where a system selectively queries the environment based on the current state of the discrimination process.
The environment is modelled as a controlled i.i.d. process conditionned by various hypotheses. Recognition is achieved when the test identifies the correct hypothesis describing the environment behavior.
As the testing proceeds, hypotheses may be rejected with infinite confidence when feature values with zero probability are observed. The sampling strategy is stationary but is updated each time a hypothesis is rejected. It is chosen according to a criterion measuring the recognition error speed of convergence to zero when the number of samples goes to infinity. This criterion is obtained by Large Deviation Theory techniques and characterizes globally the multi-hypothesis discrimination problem. An application on 2D rotation invariant shape recognition with non-closed noisy contours illustrates the approach.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
References
Breiman, L., Stone, C., Olshen, R., Friedman, J.: Classification and Regression Trees. Wadsworth, Belmont (1984)
Amit, Y., Geman, D.: Shape quantization and recognition with randomized trees. Neural Computation 9, 1545–1588 (1997)
Geman, D., Jedynak, B.: An active testing model for tracking roads in satellite images. IEEE Trans. Pattern Analysis and Machine Intelligence 18, 1–14 (1996)
Azencott, R., Vayatis, N.: Refined exponential rates in Vapnik-Chervonenkis inequalities. C. R. Acad. Sci., Paris, Math., Ser. I 332, 563–568 (2001)
Yuille, A., Coughlan, J., Wu, Y., Zhu, S.: Order parameters for detecting target curves in images: When does high level knowledge help? International Journal of Computer Vision 41, 9–33 (2001)
Wu, Y., Zhu, S., Liu, X.: Equivalence of julesz ensembles and FRAME models. International Journal of Computer Vision 38, 245–261 (2000)
Herbin, S.: Recognizing 3D objects by generating random actions. In: IEEE Conference on Computer Vision and Pattern Recognition, pp. 35–40 (1996)
Herbin, S.: Combining geometric and probabilistic structure for active recognition of 3D objects. In: Burkhardt, H., Neumann, B. (eds.) ECCV 1998. LNCS, vol. 1407, pp. 748–764. Springer, Heidelberg (1998)
Herbin, S.: Similarity measures between feature maps - application to texture comparison. In: Proceedings of the Texture 2002 workshop, pp. 67–72 (2002)
Berge, C.: Graphes et hypergraphes. Dunod (1970)
Kazakos, D.: Asymptotic error probability expressions for multihypothesis testing using multisensor data. IEEE Trans. Systems, Man and Cybernetics 21, 1101–1114 (1991)
Dembo, A., Zeitouni, O.: Large Deviations Techniques and Applications. Jones and Bartlett Publishers, Boston (1993)
Loncaric, S.: A survey of shape analysis techniques. Pattern Recognition 31, 983–1001 (1998)
Amir, A., Lindenbaum, M.: A generic grouping algorithm and its quantitative analysis. IEEE Trans. Pattern Analysis and Machine Intelligence 20, 168–185 (1998)
Olson, C., Huttenlocher, D.: Automatic target recognition by matching oriented edge pixels. IEEE Trans. Image Processing 6, 103–113 (1997)
Shimkin, N.: Extremal large deviations in controlled i.i.d. processes with applications to hypothesis testing. Adv. Appl. Probab. 25, 875–894 (1993)
Author information
Authors and Affiliations
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2003 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Herbin, S. (2003). Active Sampling Strategies for Multihypothesis Testing. In: Rangarajan, A., Figueiredo, M., Zerubia, J. (eds) Energy Minimization Methods in Computer Vision and Pattern Recognition. EMMCVPR 2003. Lecture Notes in Computer Science, vol 2683. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-540-45063-4_7
Download citation
DOI: https://doi.org/10.1007/978-3-540-45063-4_7
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-40498-9
Online ISBN: 978-3-540-45063-4
eBook Packages: Springer Book Archive