Weighted integral method for describing edges

A new method for parametrically describing edges in a given image is proposed. The method firstly obtains a set of the descriptions of edge patches in a given image by means of a weighted integral method. Employing a Gaussian function for the representation of edge patches, the weighted integral method estimates the parameter values of the Gaussian from a small local window in the given image. The estimates are computed in an algebraic way and no scan search of the parameters is required. Applying the weighted integral method at every location in the image, one obtains the set of the descriptions of edge patches, many of which overlap each other and which are redundant. The proposed method, then, greedily selects a smaller number of the patches that can describe edges in the whole part of the given image. Experimental results showed that the weighted integral method estimated the parameter values more accurately and robustly than existing methods.