An adaptive edge detection method using a modified sigmoid-LMS algorithm

An adaptive technique Of identifying edge pixcls in grey. scale images is proposed. 7hc adaptive edge detector consists of a one-dimensional Finile Impulse Response (FIR) adaplive filler followed by a maximum output sequence demtion op eram U) delecl peaks in the filler oulput. Filler coefficients for a class of images are obtained through mining on a subset of image &la using a madified Sigmoid-LMS algorithm. The iterative algorithm is shown U) converge to the linear LMS solution in cats for which E[eLsE] - 0 for suitable COOsuiwlts an the form of nonlinearity used for mining. ~n approximate expression for the bias in the weight vecm from the linw LMS solution for nonlinear system madeling Is derived. Detection results for bath synthesimd and real images arc presenled.