It is useful to ascertain whether or not the watermark is robust to operations commonly performed on images like cropping. Located below left is a 128 x 128 subregion extracted from the center of the watermarked Mona Lisa image. The watermark below right was extracted soley from this subregion. This was accomplished as follows: First, the subregion was padded with ones so that its size matched the original image. Next, the corresponding subregion in the unwatermarked image was extracted and padded also. The watermark was then computed by dividing the fourier transform of the watermarked image by the fourier transform of the unwatermarked image. As an additional noise reduction step, the extracted watermark was averaged with mirrored copies of itself. Finally, to improve contrast, it was histogram equalized.