A NEURAL NETWORK MODEL ADAPTATION FOR USE WITH UNCONSTRAINED HANDWRITTEN TEXT RECOGNITION SYSTEM

Abstract

The paper proposes an approach to adaptation of neural network models for
creating OCR-systems, designed to work with uncjnstrained handwriting. The approach is
based on the rejection of recognition of continuous characters and the transition to the
recognition of individual strokes, which are then going to the characters and / or words of
text. This approach can significantly reduce the dimension of neural networks used in the
OCR-systems that will enhance their productivity and quality of recognition.