Multi-scale Edge Detection Using Undecimated Wavelet Transform

Abstract

This paper presents multi-scale edge detection method using an undecimated Haar wavelet transform. The use of undecimated transform improves the localization of detected edges when compared to the classical, decimated Haar wavelet transform. The presented method tracks for edges that exist at several dyadic scales, favoring edges at larger scales. Edge points are obtained by non-maximum suppression in four possible directions, combined with hysteresis thresholding. The experimental results show that this method is competitive to classical edge detection methods. This multi-scale approach brings robustness to noise, while the redundancy from the undecimated transform ensures good edge localization.