A Novel Approach for Caption Detection in Videos

Since caption in videos provides much useful semantic information,caption detection methods are crucial for video analysis.A novel and efficient method is proposed to detect caption in videos.Caption detection involves three steps.First,a binary map is synthesized by an edge map and a corner map.Then the candidate caption regions are selected by horizontal projection and vertical projection.After that,several discriminative features are developed to filter out non-caption regions.The experimental results on a large volume of videos show that the proposed method is robust to various background complexities.Our method can also handle multilingual caption well.