使用复合注意力的文本图像超分辨率算法研究

Research on text image super-resolution algorithm using compound attention

现实场景下,低分辨率的文本图像会导致识别准确率降低。为了解决这个问题,提出一种使用复合注意力的文本图像超分辨率方法。在复合注意力模块中,将空间注意力模块改进为优化残差模块,并在优化残差模块中融入增强型空间注意力模块,从而重建出清晰的超分辨率文本图像。同时,引入多层训练损失函数即均方误差损失和小波损失,增加文本边缘的锐度,使生成图像中的文本具有清晰的边缘和正确的结构。实验结果证明,提出方法相较于对比方法,视觉感受方面,文字结构和边缘细节更清晰,对比度更高;客观指标方面,图像质量评价指标与文本识别准确率均得到了有效提升。

In real-world scenarios,low resolution text images can lead to a decrease in recognition accuracy.To address this issue,this article proposes a text image super-resolution method using compound attention.In the composite attention module,the spatial attention module is improved to an optimized residual module,and an enhanced spatial attention module is incorporated into the optimized residual module to reconstruct clear super-resolution text images.At the same time,the introduction of multi-layer training loss functions,namely mean square error loss and wavelet loss,increases the sharpness of text edges,resulting in clear edges and correct structure of the generated text in the image.The experimental results show that compared to the comparison method,the proposed method in this article has clearer text structure and edge details in terms of visual perception,and higher contrast.In terms of objective indicators,both image quality evaluation indicators and text recognition accuracy have been effectively improved.