Video text detection based on text edge map

Video text often contains highly useful semantic information, the analysis of which can to a great extent facilitate video retrieving and understanding. In this paper, a novel method has been proposed to detect video text. Text character strokes display dense edge features in both horizontal and vertical directions. Therefore the authors create a text edge map on the basis of the Kirsch operators. Then the authors utilize the motion detection mask to sieve the horizontal and vertical text edge map. Finally we detect the video text on the basis of text edge map. Experiments are carried out with video sources of various sizes, languages, colors, fonts, which confirm that the proposed method is robust.