Training a Single-Layer Perceptron for an Approximate Edge Detection on a Digital Image

This paper explains the development of an algorithm that approximates edge detection on a digital image. The algorithm uses an artificial neural network, trained by our implementation of the error-correction learning algorithm. In this paper, the results using our algorithm are compared to the results of the methods Canny and Sobel which are two of the widest known edge detection algorithms. The proposed algorithm was trained by four training pairs 10x10 pixels, 20x20 pixels, 50x50 pixels and 100x100 pixels. Tests were carried out using the popular Lena image, where the best pair was selected and further tests were carried out on a high resolution image and a low resolution image.