A new video text extraction method based on stroke

The video text is an important clue to infer semantic information of video. The accuracy of text extraction is crucial to the video text retrieval. Most of previous proposed methods can not completely extract the text characters when the text blocks own complicated background or when the stroke width of the text character is unpredictable. In this paper, we put forward a text extraction method, which can accurately extract video texts with complex background and various text character sizes. Experiments are conducted on TV news videos, which show that the performance of the proposed method is encouraging.