FRAME REPRESENTATIONS FOR TEXTURE SEGMENTATION

We introduce a novel method of feature extraction for texture segmenta tion that relies on multichannel wavelet frames and 2-D envelope detec tion, We describe and compare two algorithms for envelope detection ba sed on (1) the Hilbert transform and (2) zero crossings. We present cr iteria for filter selection and discuss quantitatively their effect on feature extraction. The performance of our method is demonstrated exp erimentally on samples of both natural and synthetic textures.