On Metric Skyline Processing by PM-tree

The task of similarity search in multimedia databases is usually accomplished by range or k nearest neighbor queries. However, the expressing power of these “single-example” queries fails when the user’s delicate query intent is not available as a single example. Recently, the well-known skyline operator was reused in metric similarity search as a “multi-example” query type. When applied on a multi-dimensional database (i.e., on a multi-attribute table), the traditional skyline operator selects all database objects that are not dominated by other objects. The metric skyline query adopts the skyline operator such that the multiple attributes are represented by distances (similarities) to multiple query examples. Hence, we can view the metric skyline as a set of representative database objects which are as similar to all the examples as possible and, simultaneously, are semantically distinct. In this paper we propose a technique of processing the metric skyline query by use of PM-tree, while we show that our technique significantly outperforms the original M-tree based implementation in both time and space costs. In experiments we also evaluate the partial metric skyline processing, where only a controlled number of skyline objects is retrieved.