Heterogeneous area extraction based on background suppression and adaptive clustering

Abstract

Infrared small target detection is an extremely challenging problem, especially under a complex background. Generally, targets can be easily detected by some simple and fast algorithms in the homogeneous area, but in the heterogeneous area, advanced and complicated algorithms are always needed. Therefore, heterogeneous area extraction is an important task for us to use different detection methods in different backgrounds to achieve simplifying computation while maintaining high detection performance. In this paper, a novel heterogeneous area extraction approach is proposed. Firstly, a traditional background suppression algorithm named mean filter is used to detect a group of interesting points. Then, a new adaptive clustering algorithm based on region growing is proposed to cluster the interesting points into several clusters. Finally, heterogeneous areas can be determined according to the size of cluster and the density of interesting points in the cluster. Experimental results show that our proposed method can extract heterogeneous areas of any size quickly and accurately.