A combined algorithm for video text extraction

Video text provides high-level semantic information. However, due to the complex background in video, it is of great difficulty to extract text efficiently. Although many methods hold assumptions on single feature, such as texture, connected areas etc., there are still some problems in dealing with multilingual text extraction because of its quite different appearance. In this paper, the color and edge features are used to extract the text from the video frame. In this paper, two methods are combined, called color-edge combined algorithm, to remove text background. One of the combined methods is based on the exponential changes of text color, called Transition Map model, the other one uses the text edges of different gray level image. After removing complex background, text location is determined using the vertical and horizontal projection method. This algorithm is robust to the image with multilingual text. Through extensive comparison with other approach, experimental results on a large number of video images successfully demonstrate the efficiency of this algorithm.