Automatic time stamp extraction system for home videos

Abstract

Time stamps in videos play an important role in video clustering and retrieval. For digital camcorders, the time stamp is separately stored in the video stream using some specific bytes, thus it can be accessed easily. However, for analog camcorders, the time stamp is superimposed on the video frame and mixed into the video data, thus we cannot extract it directly. Furthermore, extraction of time stamp in home videos is much harder than that of ordinary videotext due to the poor quality of the video. We propose a robust and efficient time stamp detection and recognition system for analog home videos. To deal with the special difficulties, we propose two approaches, which can remarkably improve the overall system performance. The final outputs are shot boundaries with the corresponding time indices, which are detected and recognized by the proposed system. These results can be used as metadata for further video processing.