영화에서 글자검출을 이용한 오프닝 장면의 검출

Opening shots in a movie contain information on production of the movie. These shots are independent of the story of the movie then they could be skipped if a user does not want to watch them. In this paper, we propose a method to detect opening shots. We divide sequence of frames into several shots by detecting blank frames. Then, the shots are classified as opening shots or movie shots using a SVM classifier. The feature vectors for each shot are both the ratio of the number of frames containing texts to the total number of frames and the time duration. A text detector is used to obtain the information of existence of texts in a frame. The proposed text detector localizes texts by taking MSER(Maximally Stable Extremal Region) regions as the candidates for text regions and filtering them by spatial and temporal constraints. The proposed method is straightforward and shows good performance.