Reverse Spatial Visual Top-$k$ Query

Abstract

With the wide application of mobile Internet techniques an location-based services (LBS), massive multimedia data with geo-tags has been generated and collected. In this paper, we investigate a novel type of spatial query problem, named reverse spatial visual top- $k$ query (RSVQ k ) that aims to retrieve a set of geo-images that have the query as one of the most relevant geo-images in both geographical proximity and visual similarity. Existing approaches for reverse top- $k$ queries are not suitable to address this problem because they cannot effectively process unstructured data, such as image. To this end, firstly we propose the definition of RSVQ k problem and introduce the similarity measurement. A novel hybrid index, named VR2-Tree is designed, which is a combination of visual representation of geo-image and R-Tree. Besides, an extension of VR2-Tree, called CVR2-Tree is introduced and then we discuss the calculation of lower/upper bound, and then propose the optimization technique via CVR2-Tree for further pruning. In addition, a search algorithm named RSVQ k algorithm is developed to support the efficient RSVQ k query. Comprehensive experiments are conducted on four geo-image datasets, and the results illustrate that our approach can address the RSVQ k problem effectively and efficiently.