Normalized Singular Wavelet Feature For Pattern Recognition
In the process of pattern recognition, feature descriptors are of major importance. Among various approaches of
feature representation, wavelet based feature representation in spectral domain, is of greater importance. In the representation
of wavelet features, wavelet resolution coefficients are extracted using set of filter banks and the obtained spectral bands are
used for feature extraction based on the descriptive features. However the accuracy of these feature representation purely
depends on the spectral information it reveals. The suitability of wavelet features is more effective due to its finer resolution
representation. However, it is observed that these features are very randomly distributed and finer variations were mostly
suppressed by the coefficients having high variation. In the process of pattern recognition, these finer features also hold
information, which affects the retrieval accuracy. With this objective in this paper a normalized feature for wavelet feature
extraction is proposed.
Index Terms— Wavelet features, Normalized singular features, pattern recognition.