Semantic-based Annotation Enhancement Algorithm for Semi-supervised Machine Learning Efficiency Improvement Applied to Optical Braille Recognition

The use of machine-generated annotation for unlabeled training samples, the so-called pseudo-labeling, is a known method for machine learning efficiency improvement. We propose to enforce this method by enhancement of machine-generated labeling information using processing based on semantic considerations related to the appropriate subject area. This approach was applied to the optical Braille recognition problem.