Soft computing approach to feature extraction

Based on both wavelet theory and fuzzy theory, a soft computing system (SCS) is proposed for feature extraction of signals. The proposed SCS approach possesses the advantages of soft decision-making on wavelet coefficients for feature extraction, adaptive selectivity of mapping factors to coarse-to-fine resolution, and compact form of feature representation with the SCS feature-extractor. Fuzzy sets are used to provide a robust representation for signal information, and wavelet transform is used to decompose a signal into detail and approximation signals. At a given resolution, the detail and approximation signals are inputted to the proposed SCS to extract signal features at that resolution level. The sensitivity in feature extraction of the proposed approach can be adapted by tuning the fuzzy sets for the detail and approximation signals. At different resolutions, the signal can be examined and suitable features can be extracted. Examples of both one-dimensional signals and two-dimensional fingerprint images are used to illustrate the proposed soft computing approach for feature extraction and pattern recognition. The results show that the extracted features are sensitive enough to distinguish the similar signal from the different ones and are robust enough to tolerate noise corruption.