Attention Mechanism and Representative Batch Normalization Model for Scene Text Recognition
Ling Wang, Kexin Luo, Peng Wang, Yane Bai · 2023
This paper presents a scene text recognition model, which can improve the problem of insufficient feature information extraction of irregular text images. The model uses Thin Plate Splines to correct the irregular text, and then uses ResNet to extract the feature of the text image. In order to improve the feature extraction ability of the model, the Convolutional Block Attention Module and the Representative Batch Normalization are introduced into the feature extraction module. Furthermore, experimental results show that the proposed method has an average increase of 7.8% on regular datasets and 5.4% on irregular datasets than the basic model CRNN.