Adaptive Subspace OMP for Infrared Small Target Image

This paper presents an adaptive subspace OMP method based on target attributions for reconstruction of infrared small target image. The proposed method makes use of a prior knowledge of target size and coherence of target distribution to change the structure of subspace adaptively. Firstly, the maximal projection value of redundancy vector on dictionary space is computed according to certain relevance principle. Then, according to a prior knowledge of target size and coherence of target distribution, the corresponding pixel coordinate of maximal value of projectionis set as the center point, and column stack of the surrounding pixels acts as an index set, in which the atom set could be selected. The obtained atom set is used to expand the retained one during iteration and then the subspace is obtained in this iteration, which most closely matches the redundant vector. This process changes the structure of the subspace adaptively, and more importantly, it improves the performance of reconstruction by taking advantages of OMP, SP and SAMP, which could obtain the same quality of reconstruction with less time consuming.