Scrolling text detection based on region characteristic analysis for frame rate up-conversion

Abstract

This paper proposes a new scrolling text detection method for frame rate up-conversion, which uses a text edge detector and motion vector distribution analysis of the detected text. Existing methods use either edge or motion vector information of an image for scrolling text detection. Thus, when text begins or ends scrolling at the frame boundary, they have difficulty in detecting the text. Moreover, these methods use a fixed threshold to detect text edges. This makes their detection accuracy dependent on the local characteristic of an image. To overcome these drawbacks, the proposed algorithm uses both text edge detector and motion vector distribution analysis. The proposed method consists of the following three steps. In the first step, a text map is generated by the text edge detector based on region-adaptive thresholding for luminance change. Then, an initial scrolling text map is generated by calculating the difference between the previous and current text maps. Finally, the scrolling text is refined by analyzing the region characteristic of the detected text, including density, structure, and motion vector distribution. Compared with the state-of-the-art method, the experimental results show that the proposed method increases the detection accuracy by up to 17.6%. In addition, when the proposed algorithm is used for scrolling text detection, the frame rate up-conversion algorithm can generate high quality interpolated images for the scrolling text regions.