Image interpolation techniques seek to convert low-resolution images into high-resolution ones. Conventional linear interpolation methods usually have difficulty in preserving local geometric structures. Autoregression model based interpolation methods could well exploit the dual geometry similarity between the coarse and fine scales and thus obtain better results. However, to compute the local autoregression coefficients may introduce tremendous computational complexity. In this paper, we aim to simplify this computation process by adaptively selecting the optimal interpolation filter that minimizes the autoregression energy function. The proposed scheme also makes use of the so-called integral images to reduce the computational complexity greatly and thus keeps the algorithm flexible and computationally efficient at the same time. Experimental results demonstrate that the proposed method has much less computational complexity while the visual quality is even better than the state-of-art autoregression method.