On the optimal choice of integration time-step for raster simulation of a CNN for gray level image processing

Abstract

In image processing with CNN, it is impossible to have a one-to-one mapping between the CNN hardware processors and all the pixels of the image due to hardware limitations. A raster simulator provides a solution to this problem and is capable of performing single-layer CNN simulations for any size of input image. This approach is very useful for CNN researchers to simulate hardware models, test hardware implementations of CNN, develop new templates and improve existing ones. A key issue in the raster simulation of CNN is the determination of equilibrium points whose calculation is usually carried out using complex and computationally expensive numerical integration methods. We show that the simple explicit Euler integration method, together with the optimum choice of time-step size, yields the exact equilibrium points. This leads to a drastic acceleration in the simulation allowing minimal computational effort.