Development of LSTM-Based Sentence Generation Model to Improve Recognition Performance of OCR System

In recent development of deep learning algorithms, recurrent neural net-work models that can effectively reflect dependencies between input entities and LSTM models developed from them are being used in language models. In this study, a next sentence prediction model based on LSTM was implemented with the goal of improving the sentence recognition accuracy of OCR systems. The implemented model can ensure accurate recognition by generating the next sentence using a sentence generation model and comparing its similarity with the sentence entered the OCR system in case the OCR system does not recognize it accurately due to misrecognition or loss.