Joint estimation of image rotation angle and scaling factor

Scaling and rotation are often involved in image tampering, and thus have attracted wide attention in image forensics. However, existing studies usually focus on detection or parameter estimation of a single operation, e. g., scaling or rotation. When a probe image has been both scaled and rotated, the statistical characteristics on which previous methods are based become much complicated and the parameter estimation task becomes much challenging. By analyzing the interaction between the features of both operations, a joint parameter estimation method is proposed in this paper. Firstly, some candidate estimations of the rotation angle are derived from the cyclic spectrum of the questioned image. Then, by taking the dependence between the peaks caused by rotation and those caused by scaling-then-rotation into consideration, these candidates can be further checked. Finally, both the scaling factor and the rotation angle can be derived with the proposed method. The proposed method is verified by conducting experiments on a subset of the BOSSbase dataset, achieves accuracy of 84.96% in rotation angle estimation, 78.88% in scaling factor estimation, and 77.13% in the estimation of both parameters.