A system and method are provided for registering source and target images. The method includes receiving a first source image and a first scanned image. The first scanned image is one that has been generated by scanning a printed page that has been generated by printing the first source image or a transformed first source image derived from the first source image. Locations of corners of a target image in the first scanned image are identified. With a first computed transform, the corners of the target image in the first scanned image are aligned to corners of the first source image to generate an aligned target image. Local features in the source image and aligned target image are detected. A second transform is computed to align the target image with the first source image, based on the detected local features.