Vertical Text Recognition System Based on Attention Mechanism

Abstract

This paper designs a vertical text recognition system based on the attention mechanism, which can recognize not only horizontal text but also vertical text. The system is divided into two parts: recognition and web. The recognition part includes the generation of data sets, image preprocessing and text recognition. The text recognition is divided into two processes: encoding and decoding. The web part mainly displays visually to facilitate the user experience. Because the vertical text data set is scarce, the generated data set is used here to train the neural network. The size of the training set is 750,000 images, and the test set has 75,000 images. The recognition accuracy of the test on the generated test set has reached 98%.