System and Apparatus for Automatically Assigning Hierarchical Geo-tags to Images
Original Publication Date: 2010-Jan-28
Included in the Prior Art Database: 2010-Jan-28
Image sharing websites like Flickr have greatly enhanced our understanding of the world with a large repository of timely, eye catching visual experience. State-of-the-art image search engines crawl these image sharing sites to answer diverse image queries. For example, before deciding where to spend our next holiday and building a detailed trip schedule, one may leverage on the image search engines to get extra information to aid our decision. In this scenario, it is important to have the function of organizing the huge amount of online images by well-organized geographical locations, better still if a natural hierarchy of the geo-locations can be formed. Given these, we can browse the images by locations and drill-up or drill-down the images conveniently. In another personal image management scenario, the capability of automatically assigning hierarchical geo-tags for images is also of great value. As we manically take more images to record our daily life, holiday and family, etc., tons of digital images are produced and stored. Yes, taking images are easy, pleasant and help us to keep our memory. However, managing all these images can be a heavy burden for any of us. Would you go through over 30 image albums beneath your bed or clumsily sort all images in a given period and browse them one by one? Of course not! An easier way of organizing personal images is also through geographical locations where the image was taken. One can not possibly exactly remember the time of the holiday five years ago, yet he can easily remember where he had been. So the search cost will be greatly reduced if we can associate images with places they are taken. Most online images are only manually labeled with quite noisy general words like holiday or trip, and only a small portion of these images are manually labeled with broad geographical locations which are still often not accurate enough. And it is impractical to ask the imagegrapher to annotate all possible geo-locations for his images. Even if he has infinite patience for that daunting task which is quite unlikely, he may still lack a clear and complete ontology for that task. For example, he may tag all the images taken at the Tian’an men Square with Tian’an men. However, he may not annotate further for respective images as Renmin Yingxiong Jinianbei, Renmin Dahuitang or Guojia Dajuyuan, etc. The core problem we would solve here, for the two separate yet related tasks, is automatically assigning hierarchical geo-tags for images. To do this, we should: a) Construct a hierarchy of geo-locations first b) Associate related images to the hierarchy with local content similarity.