This paper proposed a novel adaptive regularization super-resolution (SR) method. The regularization term only penalizes the low-frequency components in the image. Meanwhile it protects the high-frequency components which may represent edge, in which the penalty threshold is automatically determined based on a linear function. For the selection of regularization parameter, a logarithmic function is proposed to adaptively determine the optimum regularization parameter in each iteration step instead of a constant regularization parameter. The proposed algorithm has been tested in the synthetic visiblelight image sequence and real infrared images. Experimental results show that the proposed approach is robust and can restore image details efficiently.