Combining generalized Gaussian density and energy distribution in wavelet analysis for texture classification

Wavelet decomposition has been successfully applied to the texture classification. Several features from wavelet subbands have been extracted for classification. Of these features, energy is the most commonly used. Recent research revealed that generalized Gaussian density (GGD) outperforms the energy feature by achieving higher classification accuracy. This paper analyzes the advantage and disadvantage of these two features and proposes a scheme to combine both features for texture classification. Analysis shows that the proposed feature approximate another successfully applied histogram feature, but with much less parameters. Experiments are conducted on fingerprint verification, i.e., classifying different images that belong to the same texture class. The results show that the proposed feature effectively outperforms both energy feature and GGD feature.