Pattern recognition via Ising model with long range interactions

We present a novel approach to pattern recognition, based on mapping of images onto a multidimensional parameter space defined by a long range interaction Ising model Hamiltonian. It is shown that different character patterns occupy mutually distant points within the n-dimensional space defined by a test Hamiltonian with n terms, while similar patterns occupy nearby points. By the appropriate choice of the test Hamiltonian, the approach is adjustable to various symmetry considerations, and, as opposed to neural networks, it is not hindered by pattern storage capacity. The computational effort required for recognition of one character composed of N pixels is proportional to N32.