{"id":{"repo_id":"vilnius","oai_identifier":"oai:vu.lt:elaba:107116805"},"canonical_url":"https://search.dev.ndltd.org/etd/vilnius/oai:vu.lt:elaba:107116805","repository":{"repo_id":"vilnius","name":"Vilnius University","base_url":"https://epublications.vu.lt/oai"},"display":{"title":"Lietuvių kalbos teksto atpažinimas /","abstract":"This paper reviews convolutional and recurrent neural networks and their layers. Two models of handwritten Lithuanian text recognition are analyzed: convolutional recurrent neural network and ResNet-50. Convolutional recurrent neural network model is adapted to recognize Lithuanian words with the help of a dictionary, and the ResNet-50 model is compared with the fine-tuned version. A photo dataset of Lithuanian alphabet was created to train and test the models. The subset (Lithuanian letters “Ą”, “Č”, “Ę”, “Ė”, “Į”, “Š”, “Ų”, “Ū”, “Ž”) were synthetically generated because the found “Kaggle A-Z\" dataset did not have them. Also, an additional set of photos of Lithuanian words has been created to test already trained models. Four different neural network models were trained and tested. Their results were compared with Tesseract software. From the results of alphabet dataset it can be seen that when trying to recognize similar characters, one of them is correctly recognized less often. The recognition accuracy of this dataset compared to real-world photography dataset was higher – alphabet dataset was recognized correctly with 85.35% - 93.13%, and real-world photography dataset had 64.11% - 73.19% accuracy. This is due to real-world photographs having noise, 3D object projection in a photography and angle in which the picture was taken. For the alphabet dataset the ResNet-FT model showed highest accuracy with 97.13%, and for the real-world photography dataset the CRNN * model had best results with 73,19% which surpasses Tesseract software. Possible directions for further work is speeding up the recognition process of handwritten text, achieving greater accuracy or recognizing cursive writing.","abstract_html":"This paper reviews convolutional and recurrent neural networks and their layers. Two models of handwritten Lithuanian text recognition are analyzed: convolutional recurrent neural network and ResNet-50. Convolutional recurrent neural network model is adapted to recognize Lithuanian words with the help of a dictionary, and the ResNet-50 model is compared with the fine-tuned version. A photo dataset of Lithuanian alphabet was created to train and test the models. The subset (Lithuanian letters “Ą”, “Č”, “Ę”, “Ė”, “Į”, “Š”, “Ų”, “Ū”, “Ž”) were synthetically generated because the found “Kaggle A-Z" dataset did not have them. Also, an additional set of photos of Lithuanian words has been created to test already trained models. Four different neural network models were trained and tested. Their results were compared with Tesseract software. From the results of alphabet dataset it can be seen that when trying to recognize similar characters, one of them is correctly recognized less often. The recognition accuracy of this dataset compared to real-world photography dataset was higher – alphabet dataset was recognized correctly with 85.35% - 93.13%, and real-world photography dataset had 64.11% - 73.19% accuracy. This is due to real-world photographs having noise, 3D object projection in a photography and angle in which the picture was taken. For the alphabet dataset the ResNet-FT model showed highest accuracy with 97.13%, and for the real-world photography dataset the CRNN * model had best results with 73,19% which surpasses Tesseract software. Possible directions for further work is speeding up the recognition process of handwritten text, achieving greater accuracy or recognizing cursive writing.","abstract_has_math":false,"creators":["Karmanovas, Tadas,"],"institution":"Institutional Repository of Vilnius University","degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Raudys, Aistis"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2021,"date_issued":"2021","date_published":"2021","updated_at":"2026-07-24T05:55:44Z","subjects":[],"languages":["lit"],"rights":["info:eu-repo/semantics/openAccess"],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://repository.vu.lt/VU:ELABAETD107116805&prefLang=en_US","outbound_label":"Repository record","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Raudys, Aistis"]},{"key":"dc:creator","label":"Author","values":["Karmanovas, Tadas,"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2021"]},{"key":"dc:publisher","label":"Institution","values":["Institutional Repository of Vilnius University"]},{"key":"dc:relation","label":"Dc Relation","values":["https://epublications.vu.lt/object/elaba:107116805/107116805.pdf"]},{"key":"dc:type","label":"Dc Type","values":["info:eu-repo/semantics/bachelorThesis"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["lit"]},{"key":"dc:rights","label":"Dc Rights","values":["info:eu-repo/semantics/openAccess"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://repository.vu.lt/VU:ELABAETD107116805&prefLang=en_US"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["This paper reviews convolutional and recurrent neural networks and their layers. Two models of handwritten Lithuanian text recognition are analyzed: convolutional recurrent neural network and ResNet-50. Convolutional recurrent neural network model is adapted to recognize Lithuanian words with the help of a dictionary, and the ResNet-50 model is compared with the fine-tuned version. A photo dataset of Lithuanian alphabet was created to train and test the models. The subset (Lithuanian letters “Ą”, “Č”, “Ę”, “Ė”, “Į”, “Š”, “Ų”, “Ū”, “Ž”) were synthetically generated because the found “Kaggle A-Z\" dataset did not have them. Also, an additional set of photos of Lithuanian words has been created to test already trained models. Four different neural network models were trained and tested. Their results were compared with Tesseract software. From the results of alphabet dataset it can be seen that when trying to recognize similar characters, one of them is correctly recognized less often. The recognition accuracy of this dataset compared to real-world photography dataset was higher – alphabet dataset was recognized correctly with 85.35% - 93.13%, and real-world photography dataset had 64.11% - 73.19% accuracy. This is due to real-world photographs having noise, 3D object projection in a photography and angle in which the picture was taken. For the alphabet dataset the ResNet-FT model showed highest accuracy with 97.13%, and for the real-world photography dataset the CRNN * model had best results with 73,19% which surpasses Tesseract software. Possible directions for further work is speeding up the recognition process of handwritten text, achieving greater accuracy or recognizing cursive writing."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["Lietuvių kalbos teksto atpažinimas /","Lithuanian language optical character recognition."]}]}],"canonical_facts":{"dc:contributor":["Raudys, Aistis"],"dc:creator":["Karmanovas, Tadas,"],"dc:date":["2021"],"dc:description":["This paper reviews convolutional and recurrent neural networks and their layers. Two models of handwritten Lithuanian text recognition are analyzed: convolutional recurrent neural network and ResNet-50. Convolutional recurrent neural network model is adapted to recognize Lithuanian words with the help of a dictionary, and the ResNet-50 model is compared with the fine-tuned version. A photo dataset of Lithuanian alphabet was created to train and test the models. The subset (Lithuanian letters “Ą”, “Č”, “Ę”, “Ė”, “Į”, “Š”, “Ų”, “Ū”, “Ž”) were synthetically generated because the found “Kaggle A-Z\" dataset did not have them. Also, an additional set of photos of Lithuanian words has been created to test already trained models. Four different neural network models were trained and tested. Their results were compared with Tesseract software. From the results of alphabet dataset it can be seen that when trying to recognize similar characters, one of them is correctly recognized less often. The recognition accuracy of this dataset compared to real-world photography dataset was higher – alphabet dataset was recognized correctly with 85.35% - 93.13%, and real-world photography dataset had 64.11% - 73.19% accuracy. This is due to real-world photographs having noise, 3D object projection in a photography and angle in which the picture was taken. For the alphabet dataset the ResNet-FT model showed highest accuracy with 97.13%, and for the real-world photography dataset the CRNN * model had best results with 73,19% which surpasses Tesseract software. Possible directions for further work is speeding up the recognition process of handwritten text, achieving greater accuracy or recognizing cursive writing."],"dc:format":["application/pdf"],"dc:identifier":["https://repository.vu.lt/VU:ELABAETD107116805&prefLang=en_US"],"dc:language":["lit"],"dc:publisher":["Institutional Repository of Vilnius University"],"dc:relation":["https://epublications.vu.lt/object/elaba:107116805/107116805.pdf"],"dc:rights":["info:eu-repo/semantics/openAccess"],"dc:title":["Lietuvių kalbos teksto atpažinimas /","Lithuanian language optical character recognition."],"dc:type":["info:eu-repo/semantics/bachelorThesis"]},"updated_at":"2026-07-24T05:55:44Z"}