Regularized edge detection

It has been observed that the edge detection problem in image processing is ill-posed. The ill-posed aspect of this problem is the differentiation step, which is an explicit operation in many schemes for finding edges in images. Differentiation is very sensitive to noise and its discrete version is an ill-conditioned operation. The authors concentrate on regularizing the differentiation operation and incorporating it in an edge detection scheme. They consider both a direct implementation of Miller regularization and one involving projections onto convex sets.<>