Video corpus construction and analysis

In this paper, we propose a method of handling videos based on the construction and analysis of a video corpus containing a large volume of video data and attributes. A video corpus was constructed by coding videos of a total length of approximately 180 hours by the MPEG‐2 standard and adding attributes, such as automatically extracted characteristic values and manually entered information. These attributes include, among other things, such items as cut points, camera operation, scene change techniques, captions placed by a television opaque projector, and scaled down images. The video corpora provide tools for the input, analysis, retrieval, and visualization of data and characteristic values, allowing for structuring in a distributed environment, thus enhancing their usability. We also demonstrate the use of InfoStripes, a tool intended for the visualization of videos, whose total length can amount to several hundred hours, in the form of stripes. In conclusion, as a preliminary analysis of the video corpus, we extracted common sequences using n ‐grams of frame images and analyzed the distribution of the timing of caption text appearance as well as the probability of occurrence of attributes related to the shooting and editing techniques.