This paper describes a method for designing a filter bank for segmenting modulation features in line textures. A new version of the S-Gabor filter of Heitger et al. (1992) is developed in which the range of frequencies covered by the filter can be easily controlled. A method of tuning S-Gabor filters to the modulation features in line textures is also presented. The candidate frequency set is obtained by selecting high-energy pixels in the Fourier power spectra of the texture images. In the resulting data set, low-frequency clutter is distinguished from pixels representing modulation features using Gustavson-Kessel fuzzy clustering. The filter bank that is obtained from these processes is used to segment images of fish scales, whereby regions corresponding to slow growth are characterized by attenuation of the spatial frequency of the growth-rings.
展开▼
