Text Recognition in Persian Documents Using Self-Supervised Learning

Text is one of humankind's most significant inventions essential for communication and collaboration in modern society. Extracting text from images, especially for languages with cursive and connected scripts like Persian, presents unique challenges due to issues such as font types, word structures, and the scarcity of labeled data. This paper addresses these challenges by implementing deep learning techniques, specifically self-supervised learning with convolutional and recurrent neural networks (CNNs and RNNs). Our approach successfully mitigates the data scarcity problem, achieving an accuracy of 92.9% on the prepared dataset.