Neural network recognition of magnetic ink code line using magnetically sensed time variant signal

Abstract

This paper provides a solution unique in that the speed varies with the person swiping the check past the head, and the recognition is made without any additional timing information. Techniques applied and researched vary from neural networks to various heuristics based on the properties of the signal derived from the magnetic sense head. The neural network recognition is explored in more detail to find the most optimal solution.<>