"PERSISTENCE OF EDGE-DETECTING PROPERTIES OF A CNN TO VARIATIONS OT THE CELL OUTPUT FUNCTIONS" (WedPmPO3)

Author(s) : Oleg Kanakov (University of Nizhny Novgorod, Russia) Vladimir Shalfeev (University of Nizhny Novgorod, Russia)

Abstract : A two-dimensional model of a CNN with local couplings is considered as an image converter the input of which is the profile of initial conditions and the output is the steady-state pattern in the lattice. The properties of an edge-detecting CNN are described. The side effect of checkered pattern formation is discussed. The persistence of the edge-detecting properties of the system with respect to small perturbations of the cell output function is investigated analytically. The consequences of the replacement of the piecewise-linear cell output function by another one are studied numerically.

Menu