Local-global joint decision based clustering for airport recognition

Airport recognition plays an important role in many computer vision tasks. This paper proposes a novel method for airport recognition including the generation of clustering threshold adaptively and automatically based on local-global joint decision. The novelties of the approach include: (1) a new adaptive clustering framework we called local-global joint decision based clustering is designed instead of a fixed threshold. (2) Within the special framework, we propose a statistical probability model based on a new local feature as the preliminary decision strategy which provide local information. (3) Another part of the framework we called reconfirmation strategy contains an innovative relative similarity model that uses global information to estimate the confidence coefficient of the result from preliminary decision. Extensive experimental results demonstrate the feasibility of our approach. In addition, our method outperforms the related algorithms in terms of recognition accuracy under a variety of situations.