Automated text recognition is used in autonomous driving systems, search engines, document analysis, and manyother applications. There are many techniques to extract text information from scanned documents, but text recognitionfrom arbitrary images is a much harder task. Recently suggested deep learning approaches have demonstrated highqualityresults, but they require a huge amount of data to achieve them. The process of collecting and labelling trainingdata to train a deep learning network is costly. In this paper, we suggest an approach for automatic dataset generation fortext recognition for arbitrary languages. We use a generative adversarial network structure, which is adapted to generatereadable and clear text looking naturally on the image background. We evaluate our approach using SegLink andTextboxes++ text localization models, which were trained on examples generated by SynthText and by variations of ourmethod. The comparison showed the superiority of our method on a subset of the ICDAR 2017 dataset for English andArabic languages.
展开▼