• Multimedia synchronization for video content generation
  • Image retrieval considering co-occurence relations among people
  • Event analysis in video using spatio-temporal patterns
  • Detecting intended human objects in home videos



    Image retrieval considering co-occurence relations among people


    The recent popularity of digital cameras allows us to take a large number of images. Since people are one of the most common subjects of such images, there is an increasing need for retrieving images containing a specific person from such image collections. One way of image retrieval is to compare the specific person's face in query and faces in images in image collections. Focusing on the fact that some people such as family or friends are more likely to appear in the same image than others, we develop an image retrieval system considering not only specific person's face but also the co-occurrence relations among people.







    Event analysis in video using spatio-temporal pattern


    With growth of video sharing services, an efficient annotation of videos is strongly demanded to manage an enormous number of videos. To this end, we propose a method for automatically annotating sports or surveillance videos using spatio-temporal patterns which represent the human motions in the video. The Spatio-temporal pattern can be extracted with low computational cost, because it doesn't track an individually human. As an application of our method, we have developed a system for searching similar videos to an query video in a database of soccer videos.







    Detecting intended human objects in home videos


    In many home videos which are captured by camera persons, there are the camera persons' intended objects which are intentionally captured by the camera persons. Since the intended objects play an important role to present what a camera person wants to express in the video, the video should be processed with taking the intended objects into account in many applications such as video summarization, video retargeting, and privacy protection.

    In our research, focusing on human objects, we propose a method for detecting the intended human objects based on the observation that a camera person's intention can be reflected to how the camera person moves his/her camera, i.e., the position and area of the human objects, and the camera motion in accordance with the human objects.



    Back