Some new methods of detecting step edges in digital pictures

This note describes two operators that respond to step edges, but not to ramps. The first is similar to the digital Laplacian, but uses the max, rather than the sum, of the x and y second differences. The second uses the difference between the mean and median gray levels in a neighborhood. The outputs obtained from these operators applied to a set of test pictures are compared with each other and with the standard digital Laplacian and gradient. A third operator, which uses the distance between the center and centroid of a neighborhood as an edge value, is also briefly considered; it turns out to be equivalent to one of the standard digital approximations to the gradient.