Generic Image Classification by Web Image Mining Experiments Using a Large Number of Web Images

摘要

Thanks to the recent rapid spread of digital imaging devices, the demand for generic image recognition of various kinds of scenes becomes greater. It is, however, hard to collect various kinds of training images for recognition of various kinds of scenes so far. To solve this problem, we have proposed a generic image classification system with an automatic knowledge acquisition mechanism from the World Wide Web (WWW). We call this knowledge acquisition from WWW "Web image mining". The system gathers a large number of images related to given class keywords from the Web and classifies an unknown image into one of the classes corresponding to the class keywords using gathered images as training ones. In this report, we describe how to gather more than one thousand images per class and the experimental results of image classification by using a large number of training images.