Research approaches from given papers

[Zarathustra4]

• [x] Mixed Model OCR Training on Historical Latin Script for Out-of-the-Box Recognition and Finetuning
• [x] Handwritten Text Recognition for Historical Documents
• [x] Transformers for Historical Handwritten Text Recognition

[Zarathustra4]

The first article is not very interesting. The first and main reason is because the paper includes only printed text transcription, not hand-written. The authors have used a Calamari OCR engine and open source datasets to finetune the model.

[Zarathustra4]

The second paper is written in 2011, so the authors use old-fashioned techniques to text transcripting. But we can find useful hints about terms and practises, for example, the whole field is named Handwritten Text Recognition (HTR) and the method of transcipting words from text-lines is widely used. The paper doesn't describe deep learning approaches.

[Zarathustra4]

The third paper desribes transformer-based method. The model receive a text-line image and returns a sequence of words in it. The model has CNN backbone and transformer encoder decoder. The paper is rather proposal method than ready to use application.