The multi-scale representation of covariance matrix for planar counter over its Region of Support (ROS) was defined in the framework of B-spline scale space.The eigenvector corresponding to the largest eigenvalue of covariance matrix indicated contour tangent orientation.The multiplication of change rates of contour tangent orientation at different scales was defined as multi-scale product.A local extreme of the product was reported as corner when the value of the product exceeded a threshold.The features of contour at different scales were contained in multi-scale product which enhanced the responses of corners while suppressing the effect of noise,hence the proposed algorithm not only has good localization accuracy,but also has good robustness to noise.Experimental results demonstrate that our algorithm is rotation-invariant and insensitive to slight scale transform.