Small target detection in flir imagery using multi-scale morphological filter and kernel density estimation

A new method is proposed for detecting small targets in forward looking infrared (FLIR) imagery. Firstly, the morphological reconstruction operator is improved to remove the undesirable clutter in the background. Secondly, the image is decomposed by morphological Haar wavelet transform, and the wavelet energy image computed using the horizontal and vertical detail images is fused with the scaled image to extract candidate targets coarse-to-fine. Finally, the kernel-based tracking algorithm is adopted to eliminate the false targets which can not be tracked successfully. Experiments performed on the AMCOM FLIR data set verify the validity of the algorithm.