Multiresolution 2-dimensional edge analysis using wavelets

Abstract

The authors show that wavelet transforms can be used in an empirical way to improve edge-detected pictures. Specifically, the first derivative of a Gaussian distribution function is used as the wavelet for detecting edges. The wavelet transforms are calculated using integer scales instead of dyadic scales. This empirical approach is justified by comparing the method with an exact reconstruction process that uses wavelet transforms of the original image. The approach was demonstrated in an experiment that used a magnetic resonance image of a brain.<>