Visual graph mining for graph matching

摘要

In this study, we formulate the concept of “mining maximal-size frequent subgraphs” in the challenging domain of visual data (images and videos). In general, visual knowledge can usually be modeled as attributed relational graphs (ARGs) with local attributes representing local parts and pairwise attributes describing the spatial relationship between parts. Thus, from a practical perspective, such mining of maximal-size subgraphs can be regarded as the discovery of common objects from visual data without given annotations of object bounding boxes. From a theoretical perspective, in this study, we propose a generic definition of common subgraphs among ARGs. Many previous studies can be roughly considered as special cases of the definition. In our definition, we consider 1) variations of unary/pairwise attributes among different ARGs, 2) linkage conditions of different nodes, and 3) the learning of similarity metrics for each node. The generality of our subgraph pattern proposes great challenges to the graph-mining algorithm. We propose an approximate but efficient solution to the mining problem. We conduct five experiments to evaluate our method with different kinds of visual data, including videos and RGB/RGB-D images. These experiments demonstrate the generality of the proposed method.