Since the scene adaptive nonuniformity correction (NUC) algorithm based on neural network is easy to generate the phenomenon of "ghost" when it removes the noise in the images output by infrared imaging systems, an improved adaptive nonuniformity correction algorithm is proposed. By applying kernel regression interpolation to the neural network algorithm, the probability of "ghost" phenomenon caused by the adaptive nonuniformity algorithm is reduced effectively. The experimental results show that compared with the traditional neural network algorithm, the proposed algorithm not only can eliminate nonuniformity noise effectively, but also can restrain the generation of "ghost" phenomenon greatly under the same conditions.