A rotationally invariant two-phase scheme for corner detection

A rotationally invariant two-phase scheme is proposed for the detection of corners. In the first phase, the curvature of each point on the curve is calculated based on the coordinates of five consecutive skipped points and the assumption that these five points form a B-spline curve. The points with significant curvatures are collected as candidate corners. During the second phase, the candidate corners are verified for deviations in the global trend of the curve and the candidate corners due to quantization are removed using a self-adjusting dot-product window. Application of the proposed method on the examples given in the literature shows that it is both minimal in computation time and accurate and stable in the detected position of corners. With the proposed method, recognition of an object can be convenient, since the corners detected are rotationally invariant.