IR Target Detection via Lateral Inhibition and Singular Value Decomposition

Abstract

IR (infrared) target detection has been an important technology in the field of target search and tracking. Generally, due to the influence of IR detector noise, cloud interference and other factors, IR image is blurred, the contrast is low, and the background clutter is heavy. As a result, detecting IR targets from complex background has become a challenging task, especially when the target is small, dim and shapeless. Meanwhile, when detecting and tracking a moving IR target, the method should be able to detect both small target and area target. In this paper, an IR target detection method via LI (lateral inhibition) and SVD (singular value decomposition) is proposed. Firstly, a local structure descriptor based on SVD of gradient domain is constructed, which reflects the basic structures of the local regions of an IR image. Then, combining with the local structure descriptor, a modified LI network is established to enhance target and suppress background. Meanwhile, to calculate lateral inhibition coefficients adaptively, the direction parameters of LI network are determined according to the dominant orientations obtained from SVD. Experimental results show that compared with the typical methods, the proposed method not only can detect small and area target under complex backgrounds, but also has excellent abilities of background suppression and target enhancement.