Development of Visual Image Retrieval Technology for Internet Society

- Displays 1000 images in 3-dimensional space on PC screen -

Information-Base Functions Hitachi Laboratory
Presented May 29, 2000

Information-Base Functions Hitachi Laboratory has developed an image retrieval technology for the future which enables users to instantaneously retrieve similar images from a large image database containing hundreds of thousands of images. This technology displays thousands of candidate images extracted from a database as a continually moving island-shaped ensemble in a 3-dimensional space on a PC screen, enabling users to view this ensemble from any perspective and to retrieve required images (see photograph). This technology is expected to be essential for image retrieval in the Internet society of the 21st century where vast amounts of images can be transferred at will.

In recent years, as the performance of computer systems and their peripheral devices have improved, we can access vast amounts of images. With the proliferation of the Internet and improvement of communication infrastructures, this trend will accelerate in the 21st century, with the daily transfer and storage of huge volumes of images in databases. Thus, as the number of images which can be handled increases, an image retrieval technology is required to extract target images from vast amounts of images. However, just to extract and view the top 10 or 100 candidate images from a database as at present is not sufficiently practical. Rather, there is a need for an image retrieval technology suitable for the Internet with an easy-to-use interface to effectively extract target images from vast amounts of images.

Information-Base Functions Hitachi Laboratory has developed a similar image retrieval technology which instantaneously retrieves the top 1,000 images similar to a key image specified by the user from a database containing 200,000 images, and displays all resultant images in a virtual 3-dimensional space.

The developed retrieval technology consists of a fast search engine and a user interface. The fast search engine instantaneously searches a database containing 200,000 images based on features such as color, texture, and a rough sketch, and selects the top 1,000 images with greatest similarity. The user interface is used to display the search results in a virtual 3-dimensional space on a PC screen for viewing by the user. We have developed a visualization technique to display 1,000 retrieved images in a 3-dimensional space and a walk-through function to enable users to walk around in the space to view the images. This allows users to intuitively understand the 1,000 search results for content creation and design through effective use of large image databases.

This technology is expected to be an essential image retrieval technology for the Internet age in the 21st century where vast amounts of images are freely transferred.

This technology was introduced in TV-Tokyo Channel 12 "World Business Satellite" (23:00, May 29), Nikkei Sangyo Shimbun (page 4, May 30), Nihon Kogyo Shimbun (page 5, May 30), Denpa Shimbun (page 4, May 30), Nikkan Kogyo Shimbun (page 6, May 30), Denki Shimbun (May 30), Nikkei NET IT News (May 30), ASCII 24 News (May 30), Nikkei BizTech News (May 30), and Sky Perfect TV "JNN Newsbird" (May 29).