Local angular spectrum analysis of 2D signals

Abstract

A new local spectral characterization of 2D signals based on multichannel filter analysis is presented. The class of the employed functions, called harmonic angular (HA) filters, is derived from the concept of radial projections of local polar maps of the image. HA filters perform harmonic decomposition in an angular sense, rather than in the conventional orthogonal reference of the Gabor or wavelet approaches. This allows spectral and positional characterization of the most important image features such as edges, lines, corners, textures, etc. After a survey of the theoretical framework, the application to textured fields analysis is illustrated in some detail.<>