Edge detection using neural networks
P.J. Terry, Dung T. Vu · 2002
Neural networks can be a useful tool for edge detection. Since a neural network edge detector is a nonlinear filter, it can have a built-in thresholding capability. Thus the filtering, thresholding operation of edge detection is a natural application for neural network processing. An edge-detection neural network can be trained with backpropagation using relatively few training patterns. The most difficult part of any neural network training problem is defining the proper training set. A simple method is given for the edge detection training problem. Not only can neural networks be trained to detect edges, they can also be designed from scratch, without the necessity for training. The weights of the network can be selected to match the characteristics of your favorite linear edge detection filter. The addition of a bias, and a sigmoid non-linearity on the output produces an "engineered" neural network.>