The rapid development of the Internet provides great convenience to people’s life, but it also provides a breeding ground and convenience for the spread of harmful thoughts. Network screenshots have become a new means of information transmission, the acquisition of user viewpoint usually requires text recognition first, and then uses natural language processing for data cleaning. However, some key information may be lost in the process of language processing, resulting in data distortion. Combined with the background of information security, this paper proposes a method to locate the user viewpoint in microblog screenshots by looking for specific text areas in text images. Firstly, transfer learning of the character region perception model is performed to enhance its generalization ability on the target task, and the character level positioning of the trained character region perception model is used. Then, the single character shape is analyzed by logical reasoning, and the user viewpoint text line is recognized according to the different appearance characteristics of different characters and the similar line characteristics of the characters on the same line. Finally, the logical location results are fused with the model location results. The experimental results show that the method provides a good ability to filter user viewpoint in microblog screenshots, can effectively locate user viewpoint, and achieve the purpose of obtaining specific text areas in text images.
