Abstract:In order to be able to take full advantage of the great application potential that lies in cellular neural networks (CNNs) we need to have successful design and learning techniques as well. In almost any analogic CNN algorithm that performs an image processing task, binary CNNs play an important role. We observed that all binary CNNs reported in the literature, except for a connected component detector, exhibit monotonic dynamics. In the paper we show that the local stability of a monotonic binary CNN represen… Show more
“…It was observed in [13] that almost all the templates reported in the literature that process bipolar initial values and produce bipolar outputs introduce monotonic evolution of the cell outputs in time. The only exception found was the connected component detector (CCD) template in [14].…”
A new approach for designing a cellular nonlinear network (CNN) cell is introduced. The method can be used when the processed images are bipolar, i.e., black and white. A simple circuit realization is introduced that can be applied when building up the analog part of the cell. The cell output nonlinearity is an easily realizable positive range high gain sigmoid. Moreover, a state limited model is adopted to decrease the complexity of the design. The coefficient building blocks are simple, and also because these blocks introduce an almost error free multiplication by zero, the corresponding devices can be made relatively small due to relaxed coefficient accuracy requirements. This approach yields a very small cell area on silicon and can be used with inexpensive digital CMOS processes. Simulation results are given for both static and dynamic behavior of the proposed structure. The dynamic simulation shows very fast convergence time compared to other reported approaches for CNN very large scale integration implementation.Index Terms-Analog array processors, cellular nonlinear networks, positive range, high gain.
“…It was observed in [13] that almost all the templates reported in the literature that process bipolar initial values and produce bipolar outputs introduce monotonic evolution of the cell outputs in time. The only exception found was the connected component detector (CCD) template in [14].…”
A new approach for designing a cellular nonlinear network (CNN) cell is introduced. The method can be used when the processed images are bipolar, i.e., black and white. A simple circuit realization is introduced that can be applied when building up the analog part of the cell. The cell output nonlinearity is an easily realizable positive range high gain sigmoid. Moreover, a state limited model is adopted to decrease the complexity of the design. The coefficient building blocks are simple, and also because these blocks introduce an almost error free multiplication by zero, the corresponding devices can be made relatively small due to relaxed coefficient accuracy requirements. This approach yields a very small cell area on silicon and can be used with inexpensive digital CMOS processes. Simulation results are given for both static and dynamic behavior of the proposed structure. The dynamic simulation shows very fast convergence time compared to other reported approaches for CNN very large scale integration implementation.Index Terms-Analog array processors, cellular nonlinear networks, positive range, high gain.
SUMMARYStable cellular neural networks with binary outputs implement a non-linear mapping between sets of input and output images. Such a mapping is studied in detail. We prove two theorems: the ÿrst one yields a su cient condition in order that the non-linear mapping be well-deÿned; the second one yields a condition, that allows to describe the mapping through a simple algorithm based on the sign of the initial derivatives. Then we enunciate two additional theorems and two corollaries, that identify the class of templates satisfying the above condition: such a class is shown to be rather large and include, as particular cases, the monotonic templates, and several kinds of non-monotonic templates. Finally, a rigorous design procedure is proposed.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.