Boundary-based corner detection using eigenvalues of covariance matrices

Abstract

In this paper we present a new measure for corner detection based on the eigenvalues of the covariance matrix of boundary points over a small region of support. It avoids false alarms for superfluous corners on circular arcs. Experimental results have shown that the proposed corner detection methods using curvature measures. It has good detection and localization for curved objects in different rotations and with varying scale changes.

Dominant point detection for planar data

“…For instance, the same size of symmetric ROS for every data point was fixed up in [2] before the corner was detected. Teh and Chin [7] had pointed out this approach caused the difficulty in such a way that there is seldom any basis for selecting suitable values for the parameters to successfully determine true corner points as features describing a curve varies enormously in size and extent.…”

“…Let Eigenvalue S has been utilized as a measure for corner detection in [2]. The authors claimed that the L S data points with sharper angles have larger S than smoother ones.…”

“…Basically, the approaches can be categorized as gray-level and boundary-based [1]. Graylevel approaches directly work on gray-level images by matching corner templates or by computing gradients at edge points whereas boundary-based approaches detect corners on the boundaries of objects [2]. In this paper, boundary-based approach is our focus as the information of a shape is prominently represented by its boundary corner.…”

“…They first determine region of support, followed by computation of relative curvature-based measure by a process of non-maximum suppression. Tsai et al [2] had proposed a measure for corner detection based on the eigenvalues of the covariance matrices of boundary points over a region of support. According to [2], curvature-based methods may detect many spurious corners for an object with circular arcs of varying radii.…”

“…Tsai et al [2] had proposed a measure for corner detection based on the eigenvalues of the covariance matrices of boundary points over a region of support. According to [2], curvature-based methods may detect many spurious corners for an object with circular arcs of varying radii. Zhu et al [8] also presented an auto-corner detection based on eigenvalues of the covariance matrices of boundary points over multi-region of support.…”