HANDWRITTEN TEXT RECOGNITION OF THE REPORTS OF GOVERNORS OF THE YENISEI PROVINCE IN THE 19TH CENTURY
P.A. SHERSTNEV, K. D. Kozhin, Anna Vladimirovna Pyataeva · 2024
The text of the reports of the governors of the Yenisei Governorate have significant peculiarities, such as pre-reform language of the reports, outdated spelling norms, old letters and symbols. In addition, photos of the reports may contain various artifacts related to the peculiarities of writing and scanning of these documents. Standard recognition methods for recognizing report texts are inefficient, so the authors have developed a specialized algorithm for recognizing text in such images. The algorithm includes two stages: line detection in the image using the CRAFT model and subsequent text recognition using the TPS-ResNet-BiLSTM-Attn model. Self-created combined datasets were used for training and testing the models. Experimental studies confirm the effectiveness of the proposed methods.