<p>The appropriate image prior and regularization parameter selection are two key issues in regularization method for image restoration. In this paper, we consider a novel image regularization method based on the deep image prior (DIP) framework and total variation. By adopting a dual formulation of the TV norm, we reformulate the minimization problem as a minimax problem. A proximal point iterative method is developed to solve the model and the regularization parameter is adaptively determined by using the discrepancy principle at each iteration step. Convergence analysis of the proposed algorithm is also established. Numerical experiments demonstrate that the proposed method can achieve better image restoration quality compared to the existing approaches for image restoration with adaptive parameter selection.</p>