| With the rapid development and extensive application of computer digital image processing technology,techniques for quick modification and tampering of images have also been advancing,resulting in the proliferation of visually imperceptible manipulated images.There is an urgent and strong demand for reliable forensic techniques to determine whether digital images have been tampered with.Passive digital image forensics,as a technology that does not rely on any pre-signature extraction or pre-embedded information to authenticate the authenticity and source of an image,is becoming a new research focus in the field of multimedia security,with wide-ranging application prospects.Estimating the parameters of geometric transformations in images has become important as it can provide reliable evidence for image tampering detection.This paper focuses on passive image forensics and investigates the spectral features of scaling and rotation in digital image geometric transformations.By improving the parameter estimation algorithm,these features are applied to blind detection of image tampering,achieving excellent results.The main contributions of this paper are as follows.(1)Starting from Padin et al.’s cyclic spectrum-based algorithm for image rotation angle estimation,an improved estimator is proposed.As rotation in the spectral domain of an image often introduces noise that affects the estimation of rotation angles,this paper accurately identifies rotation peaks by aggregating one of the spectral harmonics,which was previously considered as noise in the cyclic spectrum.Experimental results show that the proposed method achieves the highest accuracy for estimating rotation angles in images rotated using bilinear and bicubic interpolation methods,with accuracies of 95.2% and 93.2% respectively.Moreover,the method demonstrates better robustness for estimating small angles and handling JPEG compression.(2)A discrimination method for different intervals of scaling factors is proposed,particularly addressing the problem of identical peak features in scaling parameters under the scale-rotation geometric transformation of an image.By using spectral normalization energy and SVM classification,these confusing scaling parameters are distinguished,achieving an overall classification accuracy of 90.5%.(3)A joint estimation method for image scaling and rotation parameters is proposed,addressing the issue of decreased accuracy in scaling factor estimation due to the mutual influence of periodic artifacts caused by the superposition of scaling and rotation operations.By considering the spatial relationship between rotation peaks and scaling peaks,the method accurately determines the peak positions to improve parameter estimation accuracy.Experimental results demonstrate that the proposed method achieves higher accuracy for dual parameter estimation in images processed using bilinear and bicubic interpolation methods,with accuracies of 93.1% and 87.6% respectively. |
