As Digital cameras and consumer-produced websites have become common, there are more chances of treating image information systems of photographs and videos. This section introduces the projects of Ito laboratory related to still and moving images.


Because digital cameras are in widespread use nowadays, more people take a large amount of photos for pleasure. In addition, images of perticular fields like medical services, securities, satellites are also tend to be stored in large quantity . We have been developing a browser which we can look through a great deal of photos at sight.
Our research product of an image browser is called CAT (Clustered Album Thumbnail) which can search images hierarchically. In CAT, images are hierarchically-divided into groups and the representative image is chosen from each group. Zooming in one of the representative image can enable you to browse all the images in the group.


Advancing our research of an image browser, we have considered its purposes for example, to search for a specific photo from so many personal pictures or to analyse one's life, treating the photos as a kind of personal history.
When we look back our lives, in many cases, we tend to associate our memories with information like 'when, where, and with who'. Following this behaviour, it will be helpful in searching for a perticular photo or analyzing a personal history.
We propose a personal photo browser 'MIAOW (Memorized Images As Organized by When/Where/Who)' with the ideas mentioned earlier. In MIAOW, photos are hierarchically devided into groups based on the locations and the date, and allocated in a 3D space, maintaining the adjacency relationships of 'when, where' information.
At the same time, the photos are also grouped by person and displayed in another window. When one photo on the first window is clicked, a corresponding person in the other window is highlighted, and when a person is clicked, the related photos are highlighted. These mutual actions make the searching much easier.


When developing the image browser, we have focused on the point that lots of meta-information and amount of characteristic can be added.
For instance, apart from the information like time and date, or the place the photo was taken, a keyword or symbolic numbers of the object on the photo (for example price, weight, evaluated value by users) often can be given.
The amount of characteristic of the image content(e.g. hue, fineness) also can be calculated. To allocate the images using those meta-information and amount of characteristic may simplify the browsing.
We have been inventing the image browser with ideas discussed above, where you can see every image with a unique feeling of the photos floating in 3D space.


To sum up the content of a long-term dynamic picture image, there is a challenge in displaying the result of summarization of the movie, and performing this task is important not only in the research of motion images, but also in the study of visualization. We have set out two types of studies as approaching this issue. The first study is to visualize the digest of human transfer pathway obtained by aerial photographs and sensors. We have proposed the visualization technique named 'FRUITE Route'. This technique can display many pieces of channel information with simplified forms, based on the consept of bundling the similar parts together in a group. The second study is the visualization which summarizes the image with lists, extracting the important frames from the motion-image sequence. We have been developing the motion-image browser which can compare the several motion-image files, aligning the keyframes in a horizontal direction and controlling the level of detail for the purpose.


The researches of an automatic recognition of the scene in images and searching its content by computer have been carried out for a few decades. However, there is an increasing need due to the popularization of image search engines.
We have continued a research on high-speed extraction of small objects appeared in a part of the image. We process the color distribution shown in a big image beforehand, which are compared with the color distribution of small objects, then extract some candidates that have similar distribution of colors, and finally choose the most appropriate part.


The research on an automatic recognition of images by computer is in progress, not only on still images but also on moving ones. One of the most well known theme is to perceive the movement or the emotion from a facial image of human. Many proposed systems estimate these movements or the emotions by detecting the displacement of characteristic points like eyes or mouth. However, it is very difficult to continue tracking the objects without losing the points.
We have produced the method which improves the tracking. This method is successful in re-tracking of the facial movement at high speed, which was once lost, based on the prediction of a movement. The comparison of a successful example and a case of failure will be helpful to see the achievement of this study.