Building a large scale visual index of landmarks and deploying a fast recognition index.

To create a fast and comprehensive landmark recognition system one has to solve many different challenges. First, information about landmarks has to be extracted from many heterogeneous sources. Second, visual models describing those landmarks need to be built and correctly matched. Finally, it is necessary to create a fast and accurate recognition engine for matching a query image against the landmark visual models. In this talk I will describe how it is possible to address the aforementioned challenges. In particular I will describe how landmark information and visual models can be constructed by mining photo sharing websites and other web resources. Also I will discuss how to efficiently search during query time such a large collection of visual information so that a real time visual recognition service can be deployed.