Previous |  Up |  Next


Mahalanobis distance; 1-D signals
This paper introduces a novel method for selecting a feature subset yielding an optimal trade-off between class separability and feature space dimensionality. We assume the following feature properties: (a) the features are ordered into a sequence, (b) robustness of the features decreases with an increasing order and (c) higher-order features supply more detailed information about the objects. We present a general algorithm how to find under those assumptions the optimal feature subset. Its performance is demonstrated experimentally in the space of moment-based descriptors of 1-D signals, which are invariant to linear filtering.
[1] Fukunaga K.: Introduction to Statistical Pattern Recognition. Academic Press, New York 1972 MR 1075415 | Zbl 0711.62052
[2] Devijver P. A., Kittler J.: Pattern Recognition: A Statistical Approach. Prentice Hall, London 1982 MR 0692767 | Zbl 0542.68071
[3] Abu–Mostafa Y. S., Psaltis D.: Recognitive aspects of moment invariants. IEEE Trans. Pattern Anal. Mach. Intell. 6 (1984), 698–706 DOI 10.1109/TPAMI.1984.4767594
[4] Teh C. H., Chin R. T.: On image analysis by the methods of moments. IEEE Trans. Pattern Anal. Mach. Intell. 10 (1988), 496–512 DOI 10.1109/34.3913 | Zbl 0709.94543
[5] Pawlak M.: On the reconstruction aspects of moment descriptors. IEEE Trans. Inform. Theory 38 (1992), 1698–1708 DOI 10.1109/18.165444 | MR 1187813 | Zbl 0761.68104
[6] Liao S. X., Pawlak M.: On image analysis by moments. IEEE Trans. Pattern Anal. Mach. Intell. 18 (1996), 254–266 DOI 10.1109/34.485554
[7] Flusser J., Suk T.: Invariants for recognition of degraded 1-D digital signals. In: Proc. 13th ICPR, Vienna 1996, vol. II, pp. 389–393
[8] Flusser J., Suk T.: Classification of degraded signals by the method of invariants. Signal Processing 60 (1997), 243–249 DOI 10.1016/S0165-1684(97)00075-3 | Zbl 1006.94512
Partner of
EuDML logo