Title: Tour the World: building a web-scale landmark recognition engine
Author:
Year: CVPR 2009
Goal: discover world-wide landmark lists and create clean visual model using different source (picasa,panoramio) , tour-guided sites ,finally, use authorship to validate the model.
[summary]
It consists of two steps:
1) get initial landmark list from (a) panoramio,picasa (get gps images)
(b) wikitravel.com (textual infomation, so need to parse the location entity from this sites)
=> Observation:
landmark is people-dependent, so the information crawled by images and travel sites may have less overlap!!
2) from the list, download the images and use image clustering techniqe to cluster image. The dense set tend to be a landmark.
[Note]
the experiment they use may be use for my research now
1. get random image out of data source to test the accuracy of our service
0 Comment(s):
Post a Comment