Optimal design of filter banks for texture discrimination

Abstract

In this paper, we present an optimum design of two-band Finite Impulse Response (FIR) Quadrature Mirror Filter bank (QMF) for maximum possible discrimination between textured images. There are several applications, which may not require reconstruction of signal from its transformed coefficients e.g. texture analysis, remote sensing etc. For such applications, features are extracted at different frequency resolution scales. Hence, it is extremely important that transformed image coefficients should be distortion less, which is not possible in practice. Therefore, we present an optimal design method for maximum possible discrimination between any particular classes of textures with minimum possible coefficient error. In order to obtain the desired results, the optimization routine adjusts relative error weighting along with passband and stopband edges for the design of symmetric response FIR filter.