Corner characterization by differential geometry techniques

An algorithm is presented for the characterization of the corner points in an image by means of three parameters of easily understandable physical meaning, namely the amplitude, aperture, and smoothness of the blurred wedge that best fits the data. Formulas are also given for the computation of the gradients and curvatures for the surface obtained convolving the step wedge of amplitude A and aperture φ with the bidimensional Gaussian function with variance σ².