Multilabel optical character recognition of Braille text at SU

This publication contains the processed Braille character image data as well as model evaluation results used in A Robust Dot-focused Classification Approach to Convolutional Braille Recognition, approved for publishing by the Journal of Universal Computer Science (J.UCS). The study was conducted as part of a BSc Computer Science masters thesis on Optical character recognition of Braille text, at Stellenbosch University (SU). The data used was sourced from two publicly available data repositories: The Angelina Set, by Ovodov, I.G. The DSBI Dataset, by Li, R., Liu, H., Wang, X. and Qian, Y. This dataset includes the train-validation-test splitting information of the Angelina set on a document level. The DSBI set was used as out-of-distribution testing set, and as such no train-test split is utilised. The dataset further contains the compressed numpy-format datasets of individual extracted and preprocessed Braille characters from each document, for the Angelina set and DSBI set separately. Two sets of model evaluation results are included, corresponding to the data used in the J.UCS article, and the expanded results reported in the Masters thesis at SU. More details on the current status of the work, as well as all future updates, models, results and source code, can be found in the related GitHub repository. The following files are included. angelina_documents_splits.tsv details the assigned subset (train / validate / test) of each document in the Angelina Braille dataset. angelina.rar contains the extracted, standardised Braille character datasets with target labels, corresponding to all characters extracted from the documents in the Angelina train / validate / test subsets. dsbi.rar contains the extracted, standardised Braille character datasets with target labels, corresponding to all characters extracted from the documents in the full DSBI dataset. Model Performances and Experiments - JUCS Article includes the primary evaluations utilised in the Journal article. Expanded Model Performances and Experiments - Stellenbosch MSc Thesis includes further evaluations with additional metrics and experiments, utilised in the Masters thesis submitted at Stellenbosch University. readme.txt contains this same brief description of the dataset.