Recurrent perceptron learning algorithm for CNNs with application to edge detection

Abstract

A supervised learning algorithm for obtaining the template coefficients in completely stable cellular neural networks (CNNs) is presented and it is applied for learning edge detection in binary images. The algorithm resembles the well-known perceptron learning algorithm and hence is called the recurrent perceptron learning algorithm (RPLA). The RPLA can be described as the following set of rules: i) Increase each feedback template coefficient which defines the connection to a mismatching cell from its whose steady-state output is the same as the mismatching cell's desired output. On the contrary, decrease each feedback template coefficient which defines the connection to a mismatching cell from its whose steady-state is different from the mismatching cell's desired output. ii) Change the input template coefficients according to the rule stated in i) by only replacing the word of neighbor with input. iii) Retain the template coefficients unchanged if the actual outputs match the desired outputs. The proposed algorithm RPLA has been applied for training CNNs to perform edge detection. The performance of the templates obtained for the chosen input-(desired)output training pairs has been tested on a set of images which are different from the input images used in the training. It has been observed that the generalization ability of CNN in edge detection is extremely well and the performance of a template found by RPLA is comparable to the performance of Canny's edge detector for two-level images.