Multi-scale corner detection based on arithmetic mean curvature

Scale-space corner detection (SSCD) has been drawing much attention in the past. Multi-scale corner detection (MSCD), which recognizes corners only at several scales, can be treated as a fast implementation of SSCD. In this paper, a new MSCD algorithm is proposed, which is based on an arithmetic mean (AM) of the k-cosine curvature values respectively computed at three scales. Compared to the existing MSCD algorithms, which are all based on a geometric mean (GM) curvature, the new algorithm yields a higher numerical stability and a lower computational cost. Experimental results have demonstrated that proposed MSCD algorithm can favorably compare with the state-of-the-art corner detection algorithms.