Online sparse IR background estimation via KRLS

Background estimation is the first step of background suppression in many infrared (IR) target detection algorithms. One sort of these algorithms consider background estimation as a supervised learning problem. On this point of view, it is necessary to search sparse solutions to control the complexity of the learned function to achieve good generalization. On the other hand, the more effective nonlinear regression algorithms are computationally demanding, so it is required to operate online. In this paper, a nonlinear online IR image background estimation algorithm based on sparse Kernel Recursive Least Squares (KRLS) is proposed. Nonlinear function regression and real IR image data experiments are performed; the results of these experiments are compared to that of original Least Squares (LS), 2-D Least Mean Squares (TDLMS) and the kernel version of LS (KLS) algorithm. The feasibility of nonlinear function regression and background estimation via this algorithm is thus demonstrated.