Large-Scale Aerial Image Categorization by Multi-Task Discriminative Topologies Discovery

Abstract

Fast and accurately categorizing the millions of aerial images on Google Maps is a useful technique in multimedia applications. Existing methods cannot handle this task effectively due to two reasons. 1) It is challenging to build a realtime image categorization system, as some geo-aware Apps update over 20 aerial images per second. 2) The aerial images' topologies are the key to distinguish their categories, but they cannot be encoded by the generic visual descriptors. To solve these two problems, we propose an efficient aerial image categorization system, aiming at mining discriminative topologies of aerial images under a multi-task learning framework. Particularly, we first construct a region adjacency graph (RAG) that describes the topology of each aerial image. Thereby, aerial image categorization can be formulated as RAG-to-RAG matching. Based on graph theory, RAG-to-RAG matching is conducted by comparing all their respective graphlets (i.e., small subgraphs). Because the number of graphlets is huge, a multi-task feature selection algorithm is derived to discover topologies jointly discriminative to multiple categories. The discovered topologies are used to extract the discriminative graphlets. Finally, these graphlets are integrated into an AdaBoost model for predicting aerial image categories. Experiments show that our approach is competitive several existing recognition models. Further, over 24 aerial images are categorized per second, reflecting that our system is ready for real-world applications.