Application of the CTPN-CRNN Model with Attention Mechanism in Scene Text Recognition

With the rapid development of the internet, textual information contained in online images has become a key resource for automated information extraction. However, this information is often embedded in images with complex backgrounds or irregular layouts, posing significant challenges to traditional text recognition methods. This paper proposes a method based on CTPN (Connectionist Text Proposal Network) and attention mechanism-enhanced CRNN (Convolutional Recurrent Neural Network) for extracting key information from online images, effectively supporting information extraction and data analysis tasks. The method first uses CTPN to accurately detect text areas in the image and generate text box candidates, then employs the CRNN enhanced with an attention mechanism for character recognition within the candidate boxes We conducted experimental validation on the ICPR MTWI 2018 dataset, which includes a variety of real-world application scenarios. The results show that the CRNN model combined with CTPN and the attention mechanism outperforms traditional OCR methods on this dataset. The high precision and robustness of this method provide strong technical support for automated information extraction, product data analysis, and intelligent search, offering broad application potential.