Multiscale estimation of multiple orientations based on morphological directional openings

Abstract : This paper introduces a novel approach to estimate multiple orientations at each pixel of a gray image at different scales. The main orientations are provided by a bank of directional openings. Gathering the responses of the filtered directional openings provide at each pixel a discrete sequence which is the directional signature. Then, the directional signature is interpolated by cubic B-splines, and the multiple orientations at each pixel are obtained by means of peak detection in the continuous directional signature. This procedure is performed using structuring elements with different lengths which results in a multiscale approach. The comparison with other existing methods as well as the experimental results on images shows the ability of the proposed method to detect multiple orientations in textured images at different scales with high accuracy.