Increasing Accuracy of Recognition Systems using Decoding Algorithms

Abstract

Handwriting recognition is a part of optical character recognition technology. Its main task is to convert handwritten text into the corresponding digital text. To accomplish this, handwriting recognition systems compare samples to identify different styles of handwriting. Although document recognition tools are quite popular, the ones for the handwritten text recognition are not accurate enough, they require a human to correct errors. This paper describes the process of recognizing lines of handwritten text in English. It also introduces modern approaches to constructing multilayer neural networks, analyzes the advantages and disadvantages of different decoding approaches and their influence on recognition accuracy as well as post-processing of output text. As a result, a neural network was developed, that recognizes lines of English handwriting up to 100 characters long and converts them into printed text.