Abstract
The recursive maximum filter (RMF) is an algorithm proposed previously to detect dim moving targets in noisy image sequences. The detection of targets is performed by thresholding the output image of the RMF. In this paper, we propose a method to determine the threshold with a constant false alarm rate (CFAR) from the output image directly. The false alarm probability depends on the distribution of the output noise. In a transient state, the distribution of the output noise changes significantly. It is, however, difficult to predict time-varying distribution because of the nonlinearity in the RMF. Consequently, it is difficult to determine the threshold in advance, and some adaptive technique is required. We show that the mean and the variance of the output noise can be estimated from the output image directly, even when there are targets. The threshold with CFAR is obtained by using the estimates.
