Пошук академічних текстів

Інформація × Реєстраційний номер 2119U006522, Матеріали видань та локальних репозитаріїв Категорія Препринт Назва роботи Development of layout analysis system for historic scholar publications Автор Bakay OlhaBakay Olha Дата публікації 01-01-2019 Постачальник інформації Український католицький університет Першоджерело https://hdl.handle.net/20.500.14570/4474 Видання Опис In this work, we compare the results of different approaches for automatic docu- ment layout analysis using Convolutional Neural Networks. Although there is great progress in the Image Processing domain, there are still open problems, such as ac- curate detection of regions of content and classification of them into semantically similar classes. The primary purpose of work is to simplify the further processing of Ukrainian historic archives. For it, two various techniques were used. The first one is modification and re-implementation of already existing approach for document lay- out analysis. Another method is suggested by us and re-uses the pre-trained model on a bigger dataset. During this work, we also collected a new dataset of Ukrainian scientific publications. We evaluate these approaches on an independent test set and compare the precisions of each model. Додано в НРАТ 2025-11-05 Закрити