An automatic tool to design CNN-UM programs

Abstract

Programs for the Cellular Neural Network - Universal Machine are usually designed explicitly, and there is no method to create them automatically. In this paper we present a tool, based on a genetic approach, capable of determining what CNN templates compose a CNN-UM program performing a given image processing task, and in which order they must be applied to the input. The effectiveness of our system is demonstrated experimentally on real-life problems, like the route number detection on public transport vehicles.