Wavelet-based Multiscale Corner Detection

摘要

A new wavelet-based multiscale corner detection algorithm for 2-D contour curves is proposed. It can successfully detect corners of different size in different scales. parametric corrdinates were introduced instead of the traditional contour orientation or curvature to maintain higher discrimination ability. A wavelet with 2 vanishing moments was used instead of traditional quadratic spline wavelet. Then the difference between real corners and noise lies in the value of local modulus maximums. So with wavelet transform's fast algorithm, this corner detection algorithm could be implemented with ease and efficiency.