An Extraction Method of Video Text in Complex Background

This paper presents a novel method to extract video caption text from background in the candidate text region. Firstly, the image of text region is binarized. Then a polarity estimation algorithm is proposed to detect the right color polarity of text. Especially, as to the hollow character, a complicated instance where text shares the same color with its background and is separated from it by stroke contours, we bring forward resolution accordingly. Finally, a connected component filtering method is employed to filter out disturbances and generate a clear binary result for recognition. Experimental results confirm that the proposed method is effective and robust.