DEVELOPMENT OF LEARNING RULES FOR HIGH-ORDER CELLULAR NEURAL NETWORKS AND APPLICABILITY IN IMAGE PROCESSING

Abstract

This paper aims at modifying a learning algorithm, developed from Recurrent Perceptron Learning Algorithm (RPLA) and a Pattern Recognition Algorithm for High-Order Cellular Neural Networks (HOCNN). Our research methods are developing theory of learning for high-order cellular neural networks and experiment with modified algorithms. The research results include two proposed algorithms and software which was built to test the two mentioned algorithms. The obtained set of the weights from our developed algorithm (named as Second-Order Recurrent Perceptron Learning Algorithm: SORPLA) can be used as filters or kernels for problems in imaging processing. In conclusion, firstly, the paper has modified the RPLA algorithm, which adds templates A and high-level templates B; secondly, it has improved the PyCNN image processing algorithm; finally, the paper also proposes an applicability of SORPLA in edge detection of image using the obtained set of the weights from the developed algorithm for the High-Order Cellular Neural Networks.