Curve and corner extraction using the multiresolution Fourier transform
Andrew Davies, Roland Wilson · Warwick Research Archive Portal (University of Warwick) · 1992
A novel method of segmenting images into regions which contain linear or circular arc features is presented. The feature models are based upon spectral properties, local estimates of which are provided, over a range of scales, by the Multiresolution Fourier Transform (MFT). The algorithm ensures that detected features both accurately model the data and are consistent across scale. A method of combining the primitive line and arc segments into more complex features is also considered. Some results from application of the algorithm to a number of natural and synthetic images are presented to illustrate its effectiveness in practical applications.