The Character Recognition Method Based on OCR
Liu Jianyang, Bao Junrong, Li Bingjin, Feng Zhiang, Su Zhang · 2023
The application scenarios of using digital documents for information exchange in modern society are gradually increasing with the rapid development of new technologies. Manually completing the classification and entry of book texts, book lending certificates and other information will also consume a lot of manpower, material resources and time costs. OCR technology based on deep learning processes data efficiently and with high accuracy which can help users better manage and process information. Therefore, this paper proposes an optical character recognition method based on deep neural network, integrating the DB Net model and the CRNN model. The main work is as follows: (1) We investigate and compare different deep learning-based text detection and text recognition algorithms. (2) Based on the features of random text position and directional orientation in text images under natural scenes, the text recognition method of DB Net and CRNN with CTC is proposed. DB Net Model outperforms other text detection models in terms of accuracy and speed and the accuracy of CRNN model can meet the requirements of real business scenarios. (3) We use the dataset to train the algorithm model and expand the method structure combined with the front-end framework. The method meets the task requirements of users to identify images and text accurately and provides a reference value for the study of the same type of methods.