Nowadays, we often use multimedia data as typified by video on a computer, so there are many researches of video data. Some of these researches have a goal immediately searching scene that user wants to watch. We need an infomation to search scene because most of the video data have no index.

We need to index based on contents of the video to access a semantic interval including a event or scene. But it is difficult for us to define semantic interval.

Video data contains the amount of characteristics called color information, we can extract this information from video data by the digital image processing technology. That is to say, we do not pass through many hands.When a place in the video changes, color changes, the contents of the video also do.

In this research, We use video shots and introduce a probability model which is used in the field of research of speech recognition to detect the semantic relation between video shots for the goal dividing video data into semantic interval.