Nowadays, due to the transition to digital data storage, there is a need to implement handwritten text recognition (HTR), which is an automatic translation of handwritten characters into a machine format. Handwriting recognition is complicated by the fact that there are many languages and it is possible to write the same character in different ways. In this regard, we conducted a study of a machine learning model for recognizing handwritten characters using databases of the Kazakh language. We trained the ResNet50 + Transformer deep learning model using two published databases of the Kazakh language: KOHTD and HKR. In the course of the study, these databases were studied on the component and qualitative sides with a comparison of the results of validation of the trained model. As a result, the KOHTD database showed results in the form of CER-9.46% and WER-20.18%, while the HKR database showed results in the form of CER-6.08% and WER-15.51%.
Keywords: ResNet50, Transformer, HTR, KOHTD, HKR, CNN, Kazakh HTR.
