A simple invariant neural network for 2-D image recognition

Abstract

A simple invariant neural network has been proposed. The network has invariance against scale and, rotation changes, in addition to the inherent shift of starting point on the image contour. This invariance comes from the new use of the MT-transform as a feature vector in a pre-processing stage. Thus, a complete invariance has been achieved, without any complexity in the network. Testing of the network, shows about a 100% recognition rate.