Orthogonal and biorthogonal texture-matched wavelet filterbanks for hierarchical texture analysis

This paper deals with the application of the wavelet transform to hierarchical texture analysis. The main goal is the design of texture-matched FIR-filterbanks as the dyadic (2-channel) orthogonal and the M-channel biorthogonal wavelet filterbanks. The new approach represents a generalization of the already known purely mathematically-oriented design methods for multirate and wavelet filterbanks. The cascaded application of the filterbanks results in a critically, by the factor M M (with M ≥ 2) downsampled, pyramidal representation of the textured image. The performance of these new filters discriminating textures is compared with the classical wavelet approaches on Brodatz ́s textures.