Text image-based super-resolution method

The invention belongs to the technical field of image processing, and provides a super-resolution method based on a text image, which comprises the following steps of: firstly, carrying out an alignment module based on a spatial transformation network, then adopting a shallow feature extraction module of a convolutional layer, and then adopting a sequence modeling module fused by a bidirectional attention network and text prior information; and then performing up-sampling on the text image, and finally forming a text image which is easy to observe. According to the method, the PBAM method is adopted, global sequence information can be extracted from the text image and can be quickly inserted into other network models, meanwhile, the quality of the text image is better recovered, the background and the edge of the text image are clearer, and the text recognition performance is better improved. 本发明属于图像处理技术领域，提供了一种基于文本图像的超分辨率方法，首先基于空间变换网络的对齐模块，接着采用卷积层的浅层特征提取模块，然后采用双向注意网络与文本先验信息融合的序列建模模块，然后再对文本图像进行上采样，最后形成易于观察的文本