Optical character recognition using a memory matrix generated from singular value decomposition

Abstract

A memory matrix provides output vectors that are specified by the corresponding input vectors. If the input vectors represent input characters, the memory matrix performs character recognition. When the input characters contain noise, it is difficult to recognize the characters. To overcome this difficulty, we generate a new memory matrix by using singular value decomposition and manipulating the singular values. The effectiveness of the memory matrix is made clear by recognizing 26 alphabets characters containing noise with an optical recognition system.