An efficient multiresolution edge detector employing a wide range of scales

Changes of image intensities can occur over a wide range of scales. Therefore, they should be analyzed accordingly. Until now, the reconstruction of a single edge map from the ones obtained at each scale has not been solved efficiently. We propose an edge detector that, using a local procedure to select an ldquooptimalrdquo scale, analyzes the input image with a virtually unlimited set of scales and yet has the same computational complexity as a single-scale convolution. The criteria to derive this algorithm and its ability to detect and localize step edges in the presence of noise are shown.