Abstract
Filterbanks are often used as preprocessing stages in many applications and can be used for e.g. feature extraction or dimensionality reduction. In this paper, an unsupervised learning algorithm is described that develops a bank of parameterized (Gabor) filters, starting from a training set. It forces the filters to partition the input space in an equitable manner: each filter is tuned to a different frequency region and contributes equally to the extraction of localized features. The algorithm is tested on a real world example, after which the results are discussed.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
References
Dony, R. D. and Haykin, S.: Optimally adaptive transform coding, IEEE Trans. on Image Processing 4(10), 1358–1370.
Field, D. J.: What is the goal of sensory coding?, Neural Computation 6 (1994), 559–601.
Gabor, D.: Theory of communication, J. IEE 93 (1946), 429–457.
Kohonen, T. K., Kaski, S. and Lappalainen, H.: Self-organized formation of various invariantfeature filters in the adaptive-subspace SOM, Neural Computation 9 (1997), 1321–1344.
Olshausen, B. A. and Field, D. J.: Emergence of simple-cell receptive field properties by learning a sparse code for natural images, Nature 381 (1996), 607–609.
TIMIT: DARPA TIMIT Acoustic-phonetic speech corpus, NIST Speech Disc 1–1.1 (CDROM), 1990.
Van Hulle, M. M. and Martinez, D.: On a novel unsupervised competitive learning algorithm for scalar quantization, IEEE Trans. on Neural Networks 5(3) (1994), 498–501.
Author information
Authors and Affiliations
Rights and permissions
About this article
Cite this article
Gautama, T., Van Hulle, M.M. Self-Organized Feature Extraction Achieved with a Parameterized Filterbank. Neural Processing Letters 10, 131–137 (1999). https://doi.org/10.1023/A:1018785519156
Issue Date:
DOI: https://doi.org/10.1023/A:1018785519156