|
The landmark recognition engine allows users to copy the URL of the image location,
then paste it into the URL of recognition engine. If the image matches a landmark image in the Google’s database,
the engine will return a results page that provide the name, location and even description of the landmark.
Previously, users could only search for photographs of a landmark
if they have either the name or description of the landmark,
but not to find the name of an unidentified landmark that they see on the Web.
The new landmark recognition engine, which is reported to have an accuracy of 80 percent,
is presented in a paper published at the recent IEEE Computer Society Conference
on Computer Vision and Pattern Recognition (CVPR09) conference in Miami.
The paper, titled
“Tour the World: Building a Web-Scale Landmark Recognition Engine”,
was co-authored by SoC PhD student Zheng Yan-Tao and his mentor Professor Chua Tat Seng,
together with their collaborators at Google.
Prof Chua and Yan-Tao are from the Lab for Media Search in NUS School of Computing. |
