Scene Location Guide by Image-Based Retrieval

In this paper, we propose a new image-based algorithm to identify where a tourist is when visiting unfamiliar places. When the tourist takes a photo of an unfamiliar place, our algorithm can recognize where the tourist is by retrieving similar images from an image database, where location information is associated with each image. Our method is not only fusing global and local information but using a coarse-to-fine three-stage search process. We first extract image descriptors from the image taken by the tourist and retrieve a number of most relevant images from the database. Then, we re-rank these relevant images based on geometric consistency. Finally, our method determines where the tourist is by using an image-to-class distance measure. Promising performance of the proposed algorithm is demonstrated by the experiments.