Fuzzy Cellular Neural Networks and Their Applications to Image Processing

• Published in: Advances in Imaging and Electron Physics Vol. 109; pp. 265 - 446
• Main Author: Yang, Tao
• Format: Book Chapter
• Language: English
• Published: Elsevier Science & Technology 1999
• ISBN: 9780120147519; 0120147513
• ISSN: 1076-5670
• DOI: 10.1016/S1076-5670(08)70199-X

This chapter discusses the theory of fuzzy cellular neural networks (FCNN) and their applications to image processing. The concepts of FCNN are reasonable extensions of CNN from classical set to fuzzy set. The principles of FCNN are based on uncertainties in human cognitive processes and in modeling neural systems. The chapter focuses on some simple cases in which only fuzzy logical OR and fuzzy logical AND are integrated. On the other hand, MAX and MIN are the simplest fuzzy union and intersection operations that can be implemented by using VLSI technologies. The structure of FCNN is a tradeoff between very large-scale integration (VLSI) implementation and general function. For the purpose of VLSI implementation, the FCNN proposed integrates the fuzzifier, the defuzzifier, and the fuzzy inference engine into a planar structure. The nonlinear dynamics of the conventional CNN are kept in FCNN structure. The chapter provides structures of type-I and type-I1 FCNNs. In a type-I FCNN, there exist fuzzy synaptic weights. The relation between a fuzzy feedback synaptic weight and an output is defined by the membership function. The relation between a fuzzy feedforward synaptic weight and an input is defined by the membership function. The inputs and outputs are crisp variables in a type-I FCNN. In a type-I1 FCNN, all synaptic weights are crisp. Inputs and outputs are supposed to be fuzzy.