Research On Key Frame Extraction Oriented To Artificial Text In Video

With the rapid development of computer technology and the popularization of digital multimedia devices,video resources become more and more abundant.In the video data,artificial text is closely related to the content of the video,and the effective detection of artificial text is of great importance for the analysis,understanding and retrieval of video and other applications.The thesis deals with shot boundary detection and key frame extraction,which are two main tasks of content-based video retrieval.Content-based video retrieval need to extract low-level visual features,complete the shot segmentation and key frame extraction according to the variation of video features between successive video frames,and then analyze and detect the image content of key frames.Because the artificial text region only occupies a small part in the whole image frame,it is difficult to determine the variation of artificial text content in the video by using traditional global-video-based methods for video frame feature extraction,which leads to the undetected key frames that contain artificial text.To solve this problem,we propose a video key frame extraction method for artificial text.In the thesis,firstly the existing methods for video boundary detection and key frame extraction are summed up;advantages and disadvantages are also discussed.Then,after study on the characteristics of video artificial text,we propose a text region localization method for video frame,in which edge and corner feature of image frame is used.Finally,by the use of pixel comparison method,the edge feature variation of the text area of a video frame is determined.Video frames with same text are classified into text sub-lens,and the last frame in the sub-len is extracted as the key frame.