A Novel Method based on Character Segmentation for Slant Chinese Screen-render Text Detection and Recognition

Screen rendering text has broad application prospects in the fields of medical records, dictionary screen capture, and screen-assisted reading. However, Chinese screen rendering text always has the challenges of small font size and low resolution. Obtaining a screen-rendered text image in a natural scene will have a certain tilt angle. These all pose great challenges for screen text recognition. This paper proposes a method based on character segmentation. In order to correct the tilt of the text image, a lightweight image text correction method is proposed. In order to obtain better character segmentation accuracy in extreme environments, a Gaussian extremum removal module is designed. Drawing on HCCR-GoogLeNet, a simple and efficient network architecture is designed for character recognition. Finally, this paper will slant text correction, horizontal text line detection, character segmentation, and character recognition to construct a complete network architecture to be applied to the screen rendering text recognition process scheme in natural scenes. The character recognition model was evaluated on ICDAR2013 and the homemade screen rendering dataset, and both achieved good results.