Abstract
Nowadays, optical character recognition (typed text recognition) and intelligent character recognition (handwritten text recognition) are applied in various life field. Many related studies have achieved great success and are highly appreciated for their applicability in study and work. However, Vietnamese characters are more various and more complex than other languages, especially the tone marks such as acute accent, grave accent, hook above, tilde, and underdot. Faced with the need for social development in Information Technology, especially the digital transformation and digitalization of the Government, documents need to be digitized and their content stored for future research and reference. Recognition and transformation scanned text is a essential problem. Although there are many tools to support the recognition of scanned text and conversion to digital versions, recognizing Vietnamese text, which is rich and diverse, including Latin characters and accents, is not easy. Vietnamese text recognition is facing a big challenge in recognizing the meaning of words. In other words, users have to manually correct spelling errors if they want to use the text, which is very time-consuming. Therefore, after optical character recognition, it is extremely necessary to develop an application that automatically corrects spelling errors in scanned documents using deep learning to complete the generation of Vietnamese text with higher accuracy than the original scan. In this thesis, we use TesseractOCR, EasyOCR and VietOCR to recognize characters in scanned documents and propose an automatic spelling correction method to get a more complete Vietnamese text than the character recognition results on the original scan. Automatic spelling correction is tested with the following deep learning networks: CNN (Convolutional Neural Network), RNN (Recurrent Neural Network), LSTM (Long Short Term Memory) to extract features.