Improved image GPS location estimation by mining salient features

Nowadays, people tend to share their personal photos, taken while they are traveling, to the social media sharing websites, such as Flickr. There is also convenient access to the large scale image dataset, usually attached with metadata such as GPS location, tags and description so on. With the help of images taken in places of interest in conjunction with the broad multimedia information realm, the task of automatic image GPS location estimation became possible. However, automatic image GPS location estimation is still a nontrivial task even in today's world with explosive quantity of images available on the website. In general, images taken from identical locations share some features, such as some salient features, even when the images are taken from different viewpoints. These salient features play an important role in the image location estimation. Thus, in this paper, we propose a salient image feature mining based image GPS location estimation method. We first mine the salient region of the input image by exploring its relation with k nearest neighboring image groups, and then select salient features by considering their relation with the neighbor image groups. Experiments on different datasets demonstrate the effectiveness of the proposed method. HighlightsA salient region mining and salient feature selection based image GPS estimation approach is proposed.We utilize the mechanism of feedback to improve GPS estimation performance. We measure the contributions of the each salient parts to GPS estimation quantitatively.We fuse saliency of each BoW in determining the image taken place.