In order to improve the performance of super-resolution restoration for an image, a new model involving general white Gaussian distributed noise and anisotropy regularization is presented. To acquire the optimal solution of the new model, immune algorithm is proposed and improved in three aspects: a memory unit group is introduced to make the algorithm perform on two independent groups parallelly; an adaptive method of exporting vaccines and inoculation is presented; and a chaos operator is implanted to the algorithm for anti-freezing. Analysis and experimental results demonstrate that the restoration based on this proposed model is robust both to different types of noises and to variances of noises. Moreover, the Improved Ratio of Signal to Noise(ISNR) of the restored image using this proposed model is 1.5 dB higher than that using traditional model. Meanwhile, improved immune algorithm can converge fastly,the steps of iteration are 8% that of the total steps for GA, and 40% that of the total steps for immune algorithm. The system consisting of the proposed novel model and improved immune algorithm is reliable on the super-resolution restoration.