Properties of multichannel texture analysis filters
Alan Conrad Bovik · International Conference on Acoustics, Speech, and Signal Processing · 2002
Texture analysis algorithms that decompose images into oriented spatial frequency channels are studied. Optimality properties for texture segmentation filters are considered using idealized (narrowband) image texture models. The functional uncertainty of the channel filters is shown to define a tradeoff between spectral selectivity and accuracy in boundary localization that is optimized by the 2-D Gabor functions. The idealized texture model is then relaxed to analyze the effects of textural perturbations interpreted as localized amplitude and phase variations on the segmentation. The effects of these perturbations are found to be effectively ameliorated with postdetection smoothing.>