DEVELOPMENT OF ALGORITHMS FOR RECOGNITION OF KAZAKH MANUSCRIPTS USING ARTIFICIAL INTELLIGENCE TECHNOLOGY

Abstract

This article discusses the use of recurrent neural networks as part of artificial intelligence for recognizing handwritten text written in the Kazakh language. Kazakh language manuscript recognition uses a method of consciously recognizing manuscripts by segmenting records and processing the results, separating the received information into text and images, and converting text to text. This is a complex structural process that involves scanning paper in one scan. In this study, A. Abdallah, M. Hamada, D. Nurseitov, in order to solve the problem of recognizing handwritten text written in Kazakh, a new model is based on fully closed convolutional neural networks and analysis of the results obtained. This paper describes a model based on the CNN-BGRU architecture (CNN - Convolutional Neural Network, Bidirectional Recursive Gate Unit, Convolutional Neural Network - Bidirectional Control Unit), and calculates error rates per character, errors per word, and handwritten sentence recognition errors. To train and test the handwriting recognition system, the Kazakh KOHTD data set was obtained. The proposed model is implemented using the TensorFlow library for Python.