|
|
Segmentation of logical units in text
Kostelník, Martin ; Kišš, Martin (oponent) ; Beneš, Karel (vedoucí práce)
The goal of this project is the topic segmentation of text into coherent units. It builds on the PERO-OCR software, aiming to improve the processing of Czech historical documents and information retrieval for librarians and scientists. This included the creation and annotation of a custom dataset comprised of 4044 pages from books, dictionaries, and periodicals. I propose an innovative approach treating segmentation as a line clustering problem. The method involves a two-stage process: initial detection of regions of interest containing text lines using the YOLOv8 model, followed by joining them using a graph neural network. This method achieves a V-measure of 77.93 %, 95.79 % and 90.23 % for books, dictionaries and periodicals, respectively.
|
|
|
Document Information Extraction
Janík, Roman ; Špaňhel, Jakub (oponent) ; Hradiš, Michal (vedoucí práce)
With development of digitization comes the need for historical document analysis. Named Entity Recognition is an important task for Information extraction and Data mining. The goal of this thesis is to develop a system for extraction of information from Czech historical documents, such as newspapers, chronicles and registry books. An information extraction system was designed, the input of which is scanned historical documents processed by the OCR algorithm. The system is based on a modified RoBERTa model. The extraction of information from Czech historical documents brings challenges in the form of the need for a suitable corpus for historical Czech. The corpora Czech Named Entity Corpus (CNEC) and Czech Historical Named Entity Corpus (CHNEC) were used to train the system, together with my own created corpus. The system achieves 88.85 F1 score on CNEC and 87.19 F1 score on CHNEC, obtaining new state-of-the-art results.
|
|
|
Deep Neural Networks for Historical Document Classification
Pinkeová, Bettina ; Kohút, Jan (oponent) ; Kišš, Martin (vedoucí práce)
The aim of this work is to create a system for historical documents classification . The task is specifically about classification of documents according to the place of origin. Several systems are proposed for solving this problem, in the work. The first designed and implemented system is based on a convolutional neural network with a self-attention mechanism instead of an average pooling layer. Another system is based on the BEiT model, which is built on a visual transformer. The BEiT model was pretrained on the task of masked image modelling and subsequently trained on the given classification task. The system based on convolutional neural network achieved an accuracy of 81.6% and the system based on masked image modelling achieved an accuracy of 82.9%. The systems implemented in this work, surpassed the systems participating in the ICDAR 2021 conference in terms of success.
|
|
| |
host ::
RSS.