A Video Caption Detection Method with Edge and Corner Features

Video caption contains important clue of video content, so it plays a significant role in the content based video retrieval. In this paper, a video text detection approach which combines edge and corner point features is proposed. Firstly, the gradient map is obtained by applying Roberts edge operator on the three color components from a single video frame. Then the Harris corner point is extracted on the gradient maps. Thirdly, the candidate text regions are acquired by morphological operations. Finally, the captions are located using the proposed text features and connected component analysis. The experiment results show the proposed approach can detect captions with high speed and high accuracy.