Dim And Small Target Background Suppression Using Morphology And Bilateral Kernel Regression

Background prediction algorithms which utilizing the correlation among the pixels to predict the background are often used to detect the dim and small target in infrared imagery. But the classical background prediction methods can’t predict the background well when the background of the infrared image is complicated and contains lots of clutter and noise. The residual image contains much clutter and noise and this results in the low detection probability and high false alarm rate. In response to this limitation, here a background prediction method based on multi-structural elements morphological filtering and bilateral kernel regression is proposed in this paper. The test image is preprocessed by morphology filtering which is simple and effective to remove clutter and noise. We can obtain the candidate targets by this step. Then the gray values of the pixels which are the potential targets we got in the previous step are estimated using the bilateral kernel regression method. We can get the precise background model after this operation.Experiment results on infrared imagery, which are characterized by low Signal to Noise Ratio, show that our newly proposed method is effective and detect the dim and small target in complex background excellently.