Single frame infrared image small target detection via patch similarity propagation based background estimation

Abstract

The detection and separation an infrared small target in a complex and noisy scene is still a challenging task for infrared-based monitoring and recognition systems. The complex edges in the real world will increase the number of false alarms for the most existing infrared small target detection algorithms. In this paper, we alleviate this problem by using a new background estimation model, which is based on patch similarity propagation, to reduce the edge interference. First, the propagated filter is improved at the patch level for small target background estimation to preserve more background structure information. Then, the Gaussian similarity parameter between patches is adaptively calculated according to the local region statistics. Then, the small target will be effectively extracted after the target component and background component are separated. The experimental results for typical scenarios demonstrate that the proposed background estimation model can simultaneously preserve more complex edges during the background estimation and produce fewer false alarms in the target image than the other baseline methods.