Image recognition into practical use, boosted by emerging video processors - Technology Trend at CEATEC JAPAN 2008 (3)

Image recognition into practical use, boosted by emerging video processors
Technology Trend at CEATEC JAPAN 2008 (3)

　Image recognition has been featured in a wide range of products and more familiar.
Also, image recognition is one of the promising applications realized by video processing engines that accelerate the HD trend in information appliances. At CEATEC JAPAN 2008, we covered image recognition technologies commercialized and relevant fields with future possibilities for popularization.

Commercially available image recognition technology

　Applications for video processing engines include SD-to-HD upconversion, video codec, and so on, as previously reported on this website. Since image recognition involves a lot of heavy processing loads, it is suitable for video processing engines, too.

　In typical image recognition processing, images input are compensated, and then characteristics are extracted. The extracted characteristics are then compared with the standard patterns for recognition, and the results are output (Figure 1). Applications employing image recognition includes a wide range of fields such as biometrics, machine vision, remote sensing, man-machine interface, medical image recognition, and so forth.

　Depending on each application, the objects for detection or recognition vary from human faces, hands, fingerprints (dactylogram), veins, and irises, to wiring shapes of printed circuit boards, scratches on manufactured products, and so on. In image recognition, the computational load is not as demanding if the objects are only detected. On the other hand, it requires more computations to recognize and authenticate individual objects, as in biometrics, which often requires hardware with higher processing performance.

　Among a number of image recognition applications, face detection particularly attracts a lot of attention, and many manufacturers demonstrated their products at CEATEC this time also. While digital still cameras featured face detection at CEATEC last year, high-definition camcorders equipped with the latest facial detection technology were outstanding this year.

　Sony showed the latest high-definition Handycam, HDR-TG1, which the company announced this April. The HDR-TG1, thanks to the face detection, can detect up to eight human faces in either still images or videos. With actual camcorders, Sony demonstrated to the visitors that anybody could shoot beautiful videos by controlling the focus, exposure, color of skin, and so on (Figure 2 & 3).

　Hitachi, Ltd. showed off Wooo DZ-BD10H their high-definition Blu-ray camcorder that supports AVCHD, which Hitachi began shipping in this August. They emphasized Wooo DZ-BD10H features facial detection they call "Kao-pita," allowing to shoot beautiful videos even when the subject is backlit or not at the center of viewfinder (Figure 4).

　Besides digital cameras and camcorders for consumers, face detection is also being implemented in videoconferencing systems and other applications. Manufacturers, who want to differentiate their products by adding high value, are fiercely competing. Also, facial recognition, requiring more processing performance than face detection, has been already adopted or applied to security cameras and systems, entry-exit systems, and so on, and the research and development is popular for both hardware and software.

Anticipated application to human-machine interface

　It is human-machine interface (HMI) that is an anticipated application of image recognition technology in the future. Today, such information appliances as PCs, cellular phones, or DVD recorders have become more advanced with a lot of features. As a result, however, it is quite complicated and difficult to operate them. It is not unusual to use just a fraction of the product features available after all.

　There are certain needs for easy-to-use information appliances such that anybody can use them easily. However, it is often pointed out that such complicatedness and difficulty are the cause of the problem, so-called "digital divide."

　For example, as a new effort to this man-machine interface issue, Toshiba implemented their gesture interface technology (Gesture Remote-Control in their term) to their entertainment notebook PC, Qosmio, that they began shipping this summer. This exemplifies that image recognition is gradually put into practical use.

　This year, Toshiba Semiconductor Company, who last year demonstrated a DVD player on a PC controlled with gesture interface, unveiled a brand-new "healing" application, "Touch Sweet Para-chan^*2" with a reference exhibit and stage presentation in their booth (Figure 5).

　Touch Sweet Para-chan is a simple game application that a Toshiba engineer developed in a short period of time for this CEATEC, using the hand gesture recognition middleware for their stream processor, SpursEngine. During their in-booth stage presentations, Toshiba elaborated ideas and even had the booth visitors experience the game with gesture interface and score to compete in each demo session (Figure 6).

　In Japan, Fixstars, a start-up specialized in Cell/B.E. software development, released a module called CVCell that speeds OpenCV on a Cell/B.E. processor. There may be a chance that the OpenCV/CVCell technology is reused and utilized on SpursEngine.

*1: ^ Created by editorial staff by HD Processing Forum, based on "1.4 Image Recognition Elemental Technology, 1.4.1 Basic Form of Image Recogniton System," Japanese Patent Office website
*2: ^ "PARA-chan" is an iconic character of Toshiba PC & network Company and is used for PC desktop accessory software, sales promotion novelties, and so forth.