Strategies of Traditional Chinese Character Recognition in Streetscape Based on Deep Learning Networks

Text recognition is an important task for extracting information from imagery data. Scene text recognition is one of its challenging scenarios since characters may have diversified fonts or sizes, be occluded by other objects and be captured from varying angles or under different light conditions. In contrast to alpha-numerical characters, Traditional Chinese Characters (TCC) receive less attention and the large number of TCC makes it difficult to collect and label enough scene-text images. This research aims at developing a set of strategies for TCC recognition. A synthetic dataset using a variety of data augmentation methods is constructed to simulate what may happen in real scenes, including text deformations, noise adding and background changes. A segmentation-based text spotting scheme is employed to locate the areas of text-lines and single characters. The characters can be recognized by the trained model and then linked into meaningful text-lines. The experimental results show that the proposed strategies work better in recognizing TCC in streetscape, when compared with existing publicly available tools.