The widespread use of digital cameras, digital video camcorders, and the Internet has led to an increasing need to search for still and video images. Attaching keywords to images to enable searches, however, is a tedious and time-consuming process and, as often happens, this approach leads to disappointing results when the attached keywords do not match those used when conducting a search.

To resolve this issue, Canon is developing an intuitive image-retrieval system that does not require keywords, but rather searches for images based on resemblances to a specified image. Using the system, users conduct searches by first selecting a still image to serve as the basis for the search, and then setting priority levels for such characteristics as color, pattern regularity, and composition. Search results are displayed as a list and can be narrowed down to enable users to quickly locate the images they are looking for.

Canon is also developing a video-image retrieval system based on the company's still-image retrieval system. This technology has established a means of identifying video clips by detecting an appropriate number of scene changes and finding desired scenes using representative key frames of the scene. When a video clip (represented by a still image) is selected, the system displays a list of video images with similar key frames, enabling users to rapidly find specific scenes, even from among large volumes of video image data. Amid the increasing popularity of harddisk video-recording devices and digital home appliances, Canon's imagecomparison search and retrieval systems have attracted high levels of attention.

Overview of Still/Video Image Comparison-Based Retrieval Technology