Scene Text Recognition Based on Positional Relation between Closed Curves

Scene text recognition has attracted much attention in the research community. Many proposed scene text recognition methods adopt a step-by-step procedure, which includes a text extraction phase and a recognition phase. In this study, in order to eliminate the risk of text extraction error, we try to build a scene text recognition system that does not involve the text extraction phase. In our proposed method, an image is viewed as a set of closed curves, and text is extracted and recognized simultaneously from the image by combining the closed curves. Through experiments, we demonstrate the recognition accuracy of our proposed method.