In the superresolution image reconstruction, the model of Hubermarkov random field is a common regularizing operator. Aiming at the unsatisfying effect of image reconstruction caused by fixed gradient threshold in the Huber function, a superresolution reconstruction algorithm is proposed based on selfadaptive gradient threshold. The regularizing model is structured based on data item and regular item under the maximum a posteriori probability framework; the regularizing parameters are updated using the intermediate results via iterative method and can solve the selected problem of gradient threshold in the model of Hubermarkov random field. Experimental results show, the improved algorithm can select the proper regularizing parameters based on local gratitude threshold and find the optimal result, recover detailed information and eliminate noise effectively.