Abstract
NEC has developed an image recognition service “GAZIRU”, where various real-world objects are recognized and linked to information by using smartphones cameras. In this paper, we explain the background, the technologies/implementation, and the commercialized applications for two of the GAZIRU services, GAZIRU-R and GAZIRU-Z. GAZIRU-R provides large-scale image search service for 2D/3D objects such as commercial products, documents, and buildings. We developed a new binary local descriptor called BRIGHT descriptor, which enables not only robust recognition against various camera capturing conditions, but also a high-speed response of one second or less for searching one million image database. GAZIRU-Z provides a counterfeit detection (authentication) solution for various industrial parts, by simply capturing image with an ordinary smartphone. It uses a “Fingerprint of Things” identification technology which identifies industrial parts individually by using its surface irregularities.
