Multispectral infrared image for target detection based on matched filtering using tensor

The problem of infrared targets detection in heavy clutter background is unsolved well. In this paper, we proposed a matched filter method based on tensor model, which generalizes the traditional matched filter, is used to distinguish between targets and clutter using multispectral infrared image. The tensor matched filter approach is adopted to take into account both spatial and spectral information simultaneously. The 3-mode product between tensor and filter matrix is given for matched filter. This method doesn't destroy the original spatial structure of multispectral images. By considering the spatial and spectral information for the filter, the probability of detection is improved. And an improved multi-way filter for multispectral tensor is performed before tensor matched filter. The experimental results of ROC curves indicate that our algorithm is effective to improve detection performance of targets in clutter background. And the multispectral images acquired at 8-9 μm , 9-10 μm , 10-11 μm , 11-12 μm , and 8-14 μm were tested in the experiment. We also show that this novel approach is extremely efficient in applications of interest by using tensor model for multispectral IR image processing.