In order to restore the original scene of a damaged image and obtain the most realistic restoration effect as much as possible, an image inpainting method was proposed based on the reliable repairing of a shifted reference image combined with the self-repair of the target image. It was inspired by the image stitching technology, if there was a reference image that can be used, it was convenient to use the SIFT (scale invariant feature transform) algorithm and RANSAC (random sample consensus) algorithm to register the reference image and the target image, and then project it to the target image. Then the Criminisi algorithm was used to self-repair the areas that had not been repaired. The authenticity and reliability of the image restoration are higher, and the repair results are less different from the actual scene, which shows the rationality of this method.