Sparse-representation-based automatic target detection in infrared imagery

A sparse-representation-based automatic target detection approach for forward-looking infrared imagery is proposed. The sparsity is calculated by approximately expressing the pixels within a center image block as the sparsest representation of its surroundings in feature domain. This locally computed sparsity reflects center-surround difference, which is used for distinguishing targets and background clutter. The experimental results on image data show that the proposed approach performs well in IR target detection. The method can not only give out the position of targets, but also the whole shape.

► The theory sparse-representation is used to design method for target detection. ► Our sparse-representation is based on the idea of center-surrounding difference. ► The structure similarity is introduced for a stable sparse solution. ► The method can not only give out the position of targets, but also the whole shape. ► The approach works accurately in real-life application, gaining low false alarm.