spatial image search on road networks returns
images based on both their spatial proximity as well as the relevancy of image contents. Existing solutions for the top-
$$ k $$
text query are not suitable to this problem since they are not sufficiently scalable to cope with hundreds of query keywords and cannot support very large road networks. In this paper, we model the problem as a top-
$$ k $$
aggregation problem. We first propose a new separate index approach that is based on the visual vocabulary tree image index and the G-tree road network index and then propose a query processing method called an external combined algorithm(CA) method. Our experimental results demonstrate that our approach outperforms the state-of-the-art hybrid method more than one order of magnitude improvement.