Large-Scale Specific Object Recognition and Its Application to Real-World Oriented Web

面向现实世界的Web是一个由信息的输入/输出标签组成的网络。只要用相机拍下一个物体的图像，就可以检索到它的相关信息。用户可以通过拍摄物体的图像将相关信息与物体联系起来。作为要处理的对象，我们考虑各种类型，如字符、文档、3D对象和人脸。在本研究中，我们开发了一种具有以下能力的目标识别方法来实现上述Web:(a)大规模（针对文档，1亿页），快速（针对文档，27ms/查询）和准确（针对文档，99%）。此外，我们还构建了一个理论框架以及面向现实世界的Web原型。

The real-world oriented Web is a web of things that play a role of input/output tags of information. Information relevant to an object can be retrieved by just taking an image of it by using a camera. The user can associate relevant information to an object by taking its image. As objects to be handled, we consider various types such as characters, documents, 3D objects and faces. In this research we develop a method of object recognition with the following capabilities for realizing the above Web: (a) large-scale (for documents, 100 million pages), fast (for documents, 27ms/query), and accurate (for documents 99%). In addition we build a theoretical framework as well as prototypes of the real-world oriented Web.