E cient Rotation Invariant Feature Extraction for Texture Segmentation - via Multiscale Wavelet Frames

This work presents an approach to the extraction of rotation invariant features for texture segmentation using multiscale wavelet frame analysis. The texture is decomposed into a set of bandpass channels by a circularly symmetric wavelet lter, which then gives a measure of edge magnitudes of the texture at di erent scales. The texture is characterized by local energies over small overlapping windows around each pixel at di erent scales. This gives features that are rotation invariant and describe the scale-space signature of the texture. A simple clustering algorithm is applied to this signature to achieve the desired segmentation. IndexFeature extraction, texture segmentation, wavelet transform, wavelet frames, non-separable lters.