Recognition of Handwritten Azerbaijani Letters using Convolutional Neural Networks

Abstract

Technology advancements have made it possible to fill out documents such as petitions and forms electronically. However, in some circumstances, hard copies of documents that are difficult to share, store, and save due to their rigid dimensions are still used to preserve documents in the conventional manner. It is crucial to convert these written documents into digital media because of this. From this view point, this goal of this study is to investigate various methods for the digitalization of handwritten documents. In this study, image processing methods were used to pre-process the documents that were converted to image format. These operations include splitting the image format of the document into the lines, separating them into words and characters, and then classifying the characters. Convolutional Neural Networks, which is used for image recognition, is one of the deep learning techniques used in classification. The Extended MNIST dataset and the symbol dataset created from the pre-existing documents are used to train the model. The success rate of the generated dataset was 88.72 percent.