Applications of CNN processing by template decomposition

Abstract

High connectivity cellular neural network (CNN) templates are inherently less robust than templates of lower connectivity. However, some types of detection tasks requiring a high degree of connectivity can be decomposed and realized by an algorithmic approach, instead of a single CNN template. The processing comprises several robust template types and logical operations. The basic template type proposed for the decomposition is at an intermediate point between high-connectivity CNN template processing and processing using digital logic exclusively.