Wavelet-based corner detection using eigenvectors of covariance matrices

Abstract

The proposed approach in this paper is to detect true corners and avoid false alarms on circular arcs by using the eigenvectors of covariance matrices and one-dimensional wavelet transform (1-D WT). The 2-D boundaries of an object are initially represented by the 1-D tangent angles calculated by the eigenvectors of covariance matrix from the boundary points coordinates over a small boundary segment. Since true corners result in stronger tangent variations, 1-D WT can be utilized to decompose the 1-D tangent angles and capture the irregular angle variations. In this manner, the locations of true corners can be easily identified by comparing the 1-D WT wavelet coefficients at high-pass decomposition levels with a pre-defined threshold. Experimental results show that the proposed method is invariant to rotation and scale under appropriate image resolution and adequate region of support for covariance matrices.