Blind Estimation of Affine Transformation Using 2-D Cyclostationarity of Resampled Images

This paper presents a unified way to estimate the parameters of affine transformation in the absence of original image. With 2-D cyclostationary characterization, we analytically show that the covariance of affine transformed image is periodic with the affine transformation matrix. Based on the relationship between the affine transformation matrix and the position of resampling-caused striking peaks in the 2-D spectrum of the image's edge map, we further study how to estimate the parameters of several typical affine transformations, e.g., the scaling factor, the rotation angle, the joint scaling and rotation transformation parameters. Examples of the output of our algorithm are shown and comparative results are presented to evaluate the performance of the proposed algorithm.