A robust and efficient universal CNN cell circuit using simplicial neuro-fuzzy inferences for fast image processing

This paper introduces a highly efficient yet easy to implement mixed-signal neural circuit. The structure combines a linear adaptive element (trained with the LMS algorithm) with a nonlinear preprocessor based on a set of fuzzy membership functions. The novelty consists in defining these functions based on the theory of simplicial decomposition, which leads to a very convenient circuit realization. The cell is particularly well suited for cellular neural networks (CNN), providing highly effective nonlinear image filters.